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 VNET_DEFINE_STATIC(u_char, ts_offset_secret[32]);
237 #define V_ts_offset_secret VNET(ts_offset_secret)
239 static int tcp_default_fb_init(struct tcpcb *tp);
240 static void tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged);
241 static int tcp_default_handoff_ok(struct tcpcb *tp);
242 static struct inpcb *tcp_notify(struct inpcb *, int);
243 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
244 static void tcp_mtudisc(struct inpcb *, int);
245 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
246 void *ip4hdr, const void *ip6hdr);
249 static struct tcp_function_block tcp_def_funcblk = {
250 .tfb_tcp_block_name = "freebsd",
251 .tfb_tcp_output = tcp_output,
252 .tfb_tcp_do_segment = tcp_do_segment,
253 .tfb_tcp_ctloutput = tcp_default_ctloutput,
254 .tfb_tcp_handoff_ok = tcp_default_handoff_ok,
255 .tfb_tcp_fb_init = tcp_default_fb_init,
256 .tfb_tcp_fb_fini = tcp_default_fb_fini,
259 int t_functions_inited = 0;
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;
265 init_tcp_functions(void)
267 if (t_functions_inited == 0) {
268 TAILQ_INIT(&t_functions);
269 rw_init_flags(&tcp_function_lock, "tcp_func_lock" , 0);
270 t_functions_inited = 1;
274 static struct tcp_function_block *
275 find_tcp_functions_locked(struct tcp_function_set *fs)
277 struct tcp_function *f;
278 struct tcp_function_block *blk=NULL;
280 TAILQ_FOREACH(f, &t_functions, tf_next) {
281 if (strcmp(f->tf_name, fs->function_set_name) == 0) {
289 static struct tcp_function_block *
290 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
292 struct tcp_function_block *rblk=NULL;
293 struct tcp_function *f;
295 TAILQ_FOREACH(f, &t_functions, tf_next) {
296 if (f->tf_fb == blk) {
307 struct tcp_function_block *
308 find_and_ref_tcp_functions(struct tcp_function_set *fs)
310 struct tcp_function_block *blk;
312 rw_rlock(&tcp_function_lock);
313 blk = find_tcp_functions_locked(fs);
315 refcount_acquire(&blk->tfb_refcnt);
316 rw_runlock(&tcp_function_lock);
320 struct tcp_function_block *
321 find_and_ref_tcp_fb(struct tcp_function_block *blk)
323 struct tcp_function_block *rblk;
325 rw_rlock(&tcp_function_lock);
326 rblk = find_tcp_fb_locked(blk, NULL);
328 refcount_acquire(&rblk->tfb_refcnt);
329 rw_runlock(&tcp_function_lock);
333 static struct tcp_function_block *
334 find_and_ref_tcp_default_fb(void)
336 struct tcp_function_block *rblk;
338 rw_rlock(&tcp_function_lock);
339 rblk = tcp_func_set_ptr;
340 refcount_acquire(&rblk->tfb_refcnt);
341 rw_runlock(&tcp_function_lock);
346 tcp_switch_back_to_default(struct tcpcb *tp)
348 struct tcp_function_block *tfb;
350 KASSERT(tp->t_fb != &tcp_def_funcblk,
351 ("%s: called by the built-in default stack", __func__));
354 * Release the old stack. This function will either find a new one
357 if (tp->t_fb->tfb_tcp_fb_fini != NULL)
358 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
359 refcount_release(&tp->t_fb->tfb_refcnt);
362 * Now, we'll find a new function block to use.
363 * Start by trying the current user-selected
364 * default, unless this stack is the user-selected
367 tfb = find_and_ref_tcp_default_fb();
368 if (tfb == tp->t_fb) {
369 refcount_release(&tfb->tfb_refcnt);
372 /* Does the stack accept this connection? */
373 if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL &&
374 (*tfb->tfb_tcp_handoff_ok)(tp)) {
375 refcount_release(&tfb->tfb_refcnt);
378 /* Try to use that stack. */
380 /* Initialize the new stack. If it succeeds, we are done. */
382 if (tp->t_fb->tfb_tcp_fb_init == NULL ||
383 (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0)
387 * Initialization failed. Release the reference count on
390 refcount_release(&tfb->tfb_refcnt);
394 * If that wasn't feasible, use the built-in default
395 * stack which is not allowed to reject anyone.
397 tfb = find_and_ref_tcp_fb(&tcp_def_funcblk);
399 /* there always should be a default */
400 panic("Can't refer to tcp_def_funcblk");
402 if (tfb->tfb_tcp_handoff_ok != NULL) {
403 if ((*tfb->tfb_tcp_handoff_ok) (tp)) {
404 /* The default stack cannot say no */
405 panic("Default stack rejects a new session?");
409 if (tp->t_fb->tfb_tcp_fb_init != NULL &&
410 (*tp->t_fb->tfb_tcp_fb_init)(tp)) {
411 /* The default stack cannot fail */
412 panic("Default stack initialization failed");
417 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
420 struct tcp_function_set fs;
421 struct tcp_function_block *blk;
423 memset(&fs, 0, sizeof(fs));
424 rw_rlock(&tcp_function_lock);
425 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
428 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
429 fs.pcbcnt = blk->tfb_refcnt;
431 rw_runlock(&tcp_function_lock);
432 error = sysctl_handle_string(oidp, fs.function_set_name,
433 sizeof(fs.function_set_name), req);
435 /* Check for error or no change */
436 if (error != 0 || req->newptr == NULL)
439 rw_wlock(&tcp_function_lock);
440 blk = find_tcp_functions_locked(&fs);
442 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
446 tcp_func_set_ptr = blk;
448 rw_wunlock(&tcp_function_lock);
452 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
453 CTLTYPE_STRING | CTLFLAG_RW,
454 NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
455 "Set/get the default TCP functions");
458 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
460 int error, cnt, linesz;
461 struct tcp_function *f;
467 rw_rlock(&tcp_function_lock);
468 TAILQ_FOREACH(f, &t_functions, tf_next) {
471 rw_runlock(&tcp_function_lock);
473 bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
474 buffer = malloc(bufsz, M_TEMP, M_WAITOK);
479 linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
480 "Alias", "PCB count");
485 rw_rlock(&tcp_function_lock);
486 TAILQ_FOREACH(f, &t_functions, tf_next) {
487 alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
488 linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
489 f->tf_fb->tfb_tcp_block_name,
490 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
491 alias ? f->tf_name : "-",
492 f->tf_fb->tfb_refcnt);
493 if (linesz >= bufsz) {
501 rw_runlock(&tcp_function_lock);
503 error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
504 free(buffer, M_TEMP);
508 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
509 CTLTYPE_STRING|CTLFLAG_RD,
510 NULL, 0, sysctl_net_inet_list_available, "A",
511 "list available TCP Function sets");
514 * Exports one (struct tcp_function_info) for each alias/name.
517 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS)
520 struct tcp_function *f;
521 struct tcp_function_info tfi;
524 * We don't allow writes.
526 if (req->newptr != NULL)
530 * Wire the old buffer so we can directly copy the functions to
531 * user space without dropping the lock.
533 if (req->oldptr != NULL) {
534 error = sysctl_wire_old_buffer(req, 0);
540 * Walk the list and copy out matching entries. If INVARIANTS
541 * is compiled in, also walk the list to verify the length of
542 * the list matches what we have recorded.
544 rw_rlock(&tcp_function_lock);
548 if (req->oldptr == NULL) {
553 TAILQ_FOREACH(f, &t_functions, tf_next) {
557 if (req->oldptr != NULL) {
558 tfi.tfi_refcnt = f->tf_fb->tfb_refcnt;
559 tfi.tfi_id = f->tf_fb->tfb_id;
560 (void)strncpy(tfi.tfi_alias, f->tf_name,
561 TCP_FUNCTION_NAME_LEN_MAX);
562 tfi.tfi_alias[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
563 (void)strncpy(tfi.tfi_name,
564 f->tf_fb->tfb_tcp_block_name,
565 TCP_FUNCTION_NAME_LEN_MAX);
566 tfi.tfi_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
567 error = SYSCTL_OUT(req, &tfi, sizeof(tfi));
569 * Don't stop on error, as that is the
570 * mechanism we use to accumulate length
571 * information if the buffer was too short.
575 KASSERT(cnt == tcp_fb_cnt,
576 ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt));
580 rw_runlock(&tcp_function_lock);
581 if (req->oldptr == NULL)
582 error = SYSCTL_OUT(req, NULL,
583 (cnt + 1) * sizeof(struct tcp_function_info));
588 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info,
589 CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE,
590 NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info",
591 "List TCP function block name-to-ID mappings");
594 * tfb_tcp_handoff_ok() function for the default stack.
595 * Note that we'll basically try to take all comers.
598 tcp_default_handoff_ok(struct tcpcb *tp)
605 * tfb_tcp_fb_init() function for the default stack.
607 * This handles making sure we have appropriate timers set if you are
608 * transitioning a socket that has some amount of setup done.
610 * The init() fuction from the default can *never* return non-zero i.e.
611 * it is required to always succeed since it is the stack of last resort!
614 tcp_default_fb_init(struct tcpcb *tp)
619 INP_WLOCK_ASSERT(tp->t_inpcb);
621 KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT,
622 ("%s: connection %p in unexpected state %d", __func__, tp,
626 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't
627 * know what to do for unexpected states (which includes TIME_WAIT).
629 if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT)
633 * Make sure some kind of transmission timer is set if there is
636 so = tp->t_inpcb->inp_socket;
637 if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) ||
638 tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) ||
639 tcp_timer_active(tp, TT_PERSIST))) {
641 * If the session has established and it looks like it should
642 * be in the persist state, set the persist timer. Otherwise,
643 * set the retransmit timer.
645 if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 &&
646 (int32_t)(tp->snd_nxt - tp->snd_una) <
647 (int32_t)sbavail(&so->so_snd))
650 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
653 /* All non-embryonic sessions get a keepalive timer. */
654 if (!tcp_timer_active(tp, TT_KEEP))
655 tcp_timer_activate(tp, TT_KEEP,
656 TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) :
663 * tfb_tcp_fb_fini() function for the default stack.
665 * This changes state as necessary (or prudent) to prepare for another stack
666 * to assume responsibility for the connection.
669 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged)
672 INP_WLOCK_ASSERT(tp->t_inpcb);
677 * Target size of TCP PCB hash tables. Must be a power of two.
679 * Note that this can be overridden by the kernel environment
680 * variable net.inet.tcp.tcbhashsize
683 #define TCBHASHSIZE 0
688 * Callouts should be moved into struct tcp directly. They are currently
689 * separate because the tcpcb structure is exported to userland for sysctl
690 * parsing purposes, which do not know about callouts.
701 VNET_DEFINE_STATIC(uma_zone_t, tcpcb_zone);
702 #define V_tcpcb_zone VNET(tcpcb_zone)
704 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
705 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
707 static struct mtx isn_mtx;
709 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
710 #define ISN_LOCK() mtx_lock(&isn_mtx)
711 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
714 * TCP initialization.
717 tcp_zone_change(void *tag)
720 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
721 uma_zone_set_max(V_tcpcb_zone, maxsockets);
722 tcp_tw_zone_change();
726 tcp_inpcb_init(void *mem, int size, int flags)
728 struct inpcb *inp = mem;
730 INP_LOCK_INIT(inp, "inp", "tcpinp");
735 * Take a value and get the next power of 2 that doesn't overflow.
736 * Used to size the tcp_inpcb hash buckets.
739 maketcp_hashsize(int size)
745 * get the next power of 2 higher than maxsockets.
747 hashsize = 1 << fls(size);
748 /* catch overflow, and just go one power of 2 smaller */
749 if (hashsize < size) {
750 hashsize = 1 << (fls(size) - 1);
755 static volatile int next_tcp_stack_id = 1;
758 * Register a TCP function block with the name provided in the names
759 * array. (Note that this function does NOT automatically register
760 * blk->tfb_tcp_block_name as a stack name. Therefore, you should
761 * explicitly include blk->tfb_tcp_block_name in the list of names if
762 * you wish to register the stack with that name.)
764 * Either all name registrations will succeed or all will fail. If
765 * a name registration fails, the function will update the num_names
766 * argument to point to the array index of the name that encountered
769 * Returns 0 on success, or an error code on failure.
772 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
773 const char *names[], int *num_names)
775 struct tcp_function *n;
776 struct tcp_function_set fs;
779 KASSERT(names != NULL && *num_names > 0,
780 ("%s: Called with 0-length name list", __func__));
781 KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
783 if (t_functions_inited == 0) {
784 init_tcp_functions();
786 if ((blk->tfb_tcp_output == NULL) ||
787 (blk->tfb_tcp_do_segment == NULL) ||
788 (blk->tfb_tcp_ctloutput == NULL) ||
789 (strlen(blk->tfb_tcp_block_name) == 0)) {
791 * These functions are required and you
797 if (blk->tfb_tcp_timer_stop_all ||
798 blk->tfb_tcp_timer_activate ||
799 blk->tfb_tcp_timer_active ||
800 blk->tfb_tcp_timer_stop) {
802 * If you define one timer function you
803 * must have them all.
805 if ((blk->tfb_tcp_timer_stop_all == NULL) ||
806 (blk->tfb_tcp_timer_activate == NULL) ||
807 (blk->tfb_tcp_timer_active == NULL) ||
808 (blk->tfb_tcp_timer_stop == NULL)) {
814 refcount_init(&blk->tfb_refcnt, 0);
816 blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1);
817 for (i = 0; i < *num_names; i++) {
818 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
825 (void)strncpy(fs.function_set_name, names[i],
826 TCP_FUNCTION_NAME_LEN_MAX);
827 fs.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
828 rw_wlock(&tcp_function_lock);
829 if (find_tcp_functions_locked(&fs) != NULL) {
830 /* Duplicate name space not allowed */
831 rw_wunlock(&tcp_function_lock);
832 free(n, M_TCPFUNCTIONS);
836 (void)strncpy(n->tf_name, names[i], TCP_FUNCTION_NAME_LEN_MAX);
837 n->tf_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
838 TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
840 rw_wunlock(&tcp_function_lock);
846 * Deregister the names we just added. Because registration failed
847 * for names[i], we don't need to deregister that name.
850 rw_wlock(&tcp_function_lock);
852 TAILQ_FOREACH(n, &t_functions, tf_next) {
853 if (!strncmp(n->tf_name, names[i],
854 TCP_FUNCTION_NAME_LEN_MAX)) {
855 TAILQ_REMOVE(&t_functions, n, tf_next);
858 free(n, M_TCPFUNCTIONS);
863 rw_wunlock(&tcp_function_lock);
868 * Register a TCP function block using the name provided in the name
871 * Returns 0 on success, or an error code on failure.
874 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
877 const char *name_list[1];
884 name_list[0] = blk->tfb_tcp_block_name;
885 rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
890 * Register a TCP function block using the name defined in
891 * blk->tfb_tcp_block_name.
893 * Returns 0 on success, or an error code on failure.
896 register_tcp_functions(struct tcp_function_block *blk, int wait)
899 return (register_tcp_functions_as_name(blk, NULL, wait));
903 * Deregister all names associated with a function block. This
904 * functionally removes the function block from use within the system.
906 * When called with a true quiesce argument, mark the function block
907 * as being removed so no more stacks will use it and determine
908 * whether the removal would succeed.
910 * When called with a false quiesce argument, actually attempt the
913 * When called with a force argument, attempt to switch all TCBs to
914 * use the default stack instead of returning EBUSY.
916 * Returns 0 on success (or if the removal would succeed, or an error
920 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce,
923 struct tcp_function *f;
925 if (strcmp(blk->tfb_tcp_block_name, "default") == 0) {
926 /* You can't un-register the default */
929 rw_wlock(&tcp_function_lock);
930 if (blk == tcp_func_set_ptr) {
931 /* You can't free the current default */
932 rw_wunlock(&tcp_function_lock);
935 /* Mark the block so no more stacks can use it. */
936 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
938 * If TCBs are still attached to the stack, attempt to switch them
939 * to the default stack.
941 if (force && blk->tfb_refcnt) {
944 VNET_ITERATOR_DECL(vnet_iter);
946 rw_wunlock(&tcp_function_lock);
949 VNET_FOREACH(vnet_iter) {
950 CURVNET_SET(vnet_iter);
951 INP_INFO_WLOCK(&V_tcbinfo);
952 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
954 if (inp->inp_flags & INP_TIMEWAIT) {
959 if (tp == NULL || tp->t_fb != blk) {
963 tcp_switch_back_to_default(tp);
966 INP_INFO_WUNLOCK(&V_tcbinfo);
971 rw_wlock(&tcp_function_lock);
973 if (blk->tfb_refcnt) {
974 /* TCBs still attached. */
975 rw_wunlock(&tcp_function_lock);
980 rw_wunlock(&tcp_function_lock);
983 /* Remove any function names that map to this function block. */
984 while (find_tcp_fb_locked(blk, &f) != NULL) {
985 TAILQ_REMOVE(&t_functions, f, tf_next);
988 free(f, M_TCPFUNCTIONS);
990 rw_wunlock(&tcp_function_lock);
997 const char *tcbhash_tuneable;
1000 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
1003 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
1004 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1005 printf("%s: WARNING: unable to register helper hook\n", __func__);
1006 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
1007 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1008 printf("%s: WARNING: unable to register helper hook\n", __func__);
1010 hashsize = TCBHASHSIZE;
1011 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
1012 if (hashsize == 0) {
1014 * Auto tune the hash size based on maxsockets.
1015 * A perfect hash would have a 1:1 mapping
1016 * (hashsize = maxsockets) however it's been
1017 * suggested that O(2) average is better.
1019 hashsize = maketcp_hashsize(maxsockets / 4);
1021 * Our historical default is 512,
1022 * do not autotune lower than this.
1026 if (bootverbose && IS_DEFAULT_VNET(curvnet))
1027 printf("%s: %s auto tuned to %d\n", __func__,
1028 tcbhash_tuneable, hashsize);
1031 * We require a hashsize to be a power of two.
1032 * Previously if it was not a power of two we would just reset it
1033 * back to 512, which could be a nasty surprise if you did not notice
1034 * the error message.
1035 * Instead what we do is clip it to the closest power of two lower
1036 * than the specified hash value.
1038 if (!powerof2(hashsize)) {
1039 int oldhashsize = hashsize;
1041 hashsize = maketcp_hashsize(hashsize);
1042 /* prevent absurdly low value */
1045 printf("%s: WARNING: TCB hash size not a power of 2, "
1046 "clipped from %d to %d.\n", __func__, oldhashsize,
1049 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
1050 "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE);
1053 * These have to be type stable for the benefit of the timers.
1055 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
1056 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1057 uma_zone_set_max(V_tcpcb_zone, maxsockets);
1058 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
1064 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
1065 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
1066 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1068 tcp_fastopen_init();
1070 /* Skip initialization of globals for non-default instances. */
1071 if (!IS_DEFAULT_VNET(curvnet))
1074 tcp_reass_global_init();
1076 /* XXX virtualize those bellow? */
1077 tcp_delacktime = TCPTV_DELACK;
1078 tcp_keepinit = TCPTV_KEEP_INIT;
1079 tcp_keepidle = TCPTV_KEEP_IDLE;
1080 tcp_keepintvl = TCPTV_KEEPINTVL;
1081 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
1082 tcp_msl = TCPTV_MSL;
1083 tcp_rexmit_min = TCPTV_MIN;
1084 if (tcp_rexmit_min < 1)
1086 tcp_persmin = TCPTV_PERSMIN;
1087 tcp_persmax = TCPTV_PERSMAX;
1088 tcp_rexmit_slop = TCPTV_CPU_VAR;
1089 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
1090 tcp_tcbhashsize = hashsize;
1091 /* Setup the tcp function block list */
1092 init_tcp_functions();
1093 register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
1095 /* Initialize the TCP logging data. */
1098 read_random(&V_ts_offset_secret, sizeof(V_ts_offset_secret));
1100 if (tcp_soreceive_stream) {
1102 tcp_usrreqs.pru_soreceive = soreceive_stream;
1105 tcp6_usrreqs.pru_soreceive = soreceive_stream;
1110 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
1112 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
1114 if (max_protohdr < TCP_MINPROTOHDR)
1115 max_protohdr = TCP_MINPROTOHDR;
1116 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
1118 #undef TCP_MINPROTOHDR
1121 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
1122 SHUTDOWN_PRI_DEFAULT);
1123 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
1124 EVENTHANDLER_PRI_ANY);
1132 tcp_destroy(void *unused __unused)
1140 * All our processes are gone, all our sockets should be cleaned
1141 * up, which means, we should be past the tcp_discardcb() calls.
1142 * Sleep to let all tcpcb timers really disappear and cleanup.
1145 INP_LIST_RLOCK(&V_tcbinfo);
1146 n = V_tcbinfo.ipi_count;
1147 INP_LIST_RUNLOCK(&V_tcbinfo);
1150 pause("tcpdes", hz / 10);
1155 in_pcbinfo_destroy(&V_tcbinfo);
1156 /* tcp_discardcb() clears the sack_holes up. */
1157 uma_zdestroy(V_sack_hole_zone);
1158 uma_zdestroy(V_tcpcb_zone);
1161 * Cannot free the zone until all tcpcbs are released as we attach
1162 * the allocations to them.
1164 tcp_fastopen_destroy();
1167 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
1169 printf("%s: WARNING: unable to deregister helper hook "
1170 "type=%d, id=%d: error %d returned\n", __func__,
1171 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
1173 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
1175 printf("%s: WARNING: unable to deregister helper hook "
1176 "type=%d, id=%d: error %d returned\n", __func__,
1177 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
1181 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
1191 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
1192 * tcp_template used to store this data in mbufs, but we now recopy it out
1193 * of the tcpcb each time to conserve mbufs.
1196 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
1198 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
1200 INP_WLOCK_ASSERT(inp);
1203 if ((inp->inp_vflag & INP_IPV6) != 0) {
1204 struct ip6_hdr *ip6;
1206 ip6 = (struct ip6_hdr *)ip_ptr;
1207 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
1208 (inp->inp_flow & IPV6_FLOWINFO_MASK);
1209 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
1210 (IPV6_VERSION & IPV6_VERSION_MASK);
1211 ip6->ip6_nxt = IPPROTO_TCP;
1212 ip6->ip6_plen = htons(sizeof(struct tcphdr));
1213 ip6->ip6_src = inp->in6p_laddr;
1214 ip6->ip6_dst = inp->in6p_faddr;
1217 #if defined(INET6) && defined(INET)
1224 ip = (struct ip *)ip_ptr;
1225 ip->ip_v = IPVERSION;
1227 ip->ip_tos = inp->inp_ip_tos;
1231 ip->ip_ttl = inp->inp_ip_ttl;
1233 ip->ip_p = IPPROTO_TCP;
1234 ip->ip_src = inp->inp_laddr;
1235 ip->ip_dst = inp->inp_faddr;
1238 th->th_sport = inp->inp_lport;
1239 th->th_dport = inp->inp_fport;
1247 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
1251 * Create template to be used to send tcp packets on a connection.
1252 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
1253 * use for this function is in keepalives, which use tcp_respond.
1256 tcpip_maketemplate(struct inpcb *inp)
1260 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
1263 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
1268 * Send a single message to the TCP at address specified by
1269 * the given TCP/IP header. If m == NULL, then we make a copy
1270 * of the tcpiphdr at th and send directly to the addressed host.
1271 * This is used to force keep alive messages out using the TCP
1272 * template for a connection. If flags are given then we send
1273 * a message back to the TCP which originated the segment th,
1274 * and discard the mbuf containing it and any other attached mbufs.
1276 * In any case the ack and sequence number of the transmitted
1277 * segment are as specified by the parameters.
1279 * NOTE: If m != NULL, then th must point to *inside* the mbuf.
1282 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
1283 tcp_seq ack, tcp_seq seq, int flags)
1292 struct ip6_hdr *ip6;
1295 int optlen, tlen, win;
1298 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
1301 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
1308 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
1309 INP_WLOCK_ASSERT(inp);
1316 if (!(flags & TH_RST)) {
1317 win = sbspace(&inp->inp_socket->so_rcv);
1318 if (win > TCP_MAXWIN << tp->rcv_scale)
1319 win = TCP_MAXWIN << tp->rcv_scale;
1321 if ((tp->t_flags & TF_NOOPT) == 0)
1325 m = m_gethdr(M_NOWAIT, MT_DATA);
1328 m->m_data += max_linkhdr;
1331 bcopy((caddr_t)ip6, mtod(m, caddr_t),
1332 sizeof(struct ip6_hdr));
1333 ip6 = mtod(m, struct ip6_hdr *);
1334 nth = (struct tcphdr *)(ip6 + 1);
1338 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1339 ip = mtod(m, struct ip *);
1340 nth = (struct tcphdr *)(ip + 1);
1342 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1344 } else if (!M_WRITABLE(m)) {
1347 /* Can't reuse 'm', allocate a new mbuf. */
1348 n = m_gethdr(M_NOWAIT, MT_DATA);
1354 if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1360 n->m_data += max_linkhdr;
1361 /* m_len is set later */
1362 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1365 bcopy((caddr_t)ip6, mtod(n, caddr_t),
1366 sizeof(struct ip6_hdr));
1367 ip6 = mtod(n, struct ip6_hdr *);
1368 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1369 nth = (struct tcphdr *)(ip6 + 1);
1373 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1374 ip = mtod(n, struct ip *);
1375 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1376 nth = (struct tcphdr *)(ip + 1);
1378 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1379 xchg(nth->th_dport, nth->th_sport, uint16_t);
1386 * XXX MRT We inherit the FIB, which is lucky.
1390 m->m_data = (caddr_t)ipgen;
1391 /* m_len is set later */
1394 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1395 nth = (struct tcphdr *)(ip6 + 1);
1399 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1400 nth = (struct tcphdr *)(ip + 1);
1404 * this is usually a case when an extension header
1405 * exists between the IPv6 header and the
1408 nth->th_sport = th->th_sport;
1409 nth->th_dport = th->th_dport;
1411 xchg(nth->th_dport, nth->th_sport, uint16_t);
1417 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1419 #if defined(INET) && defined(INET6)
1423 tlen = sizeof (struct tcpiphdr);
1427 KASSERT(M_TRAILINGSPACE(m) >= tlen,
1428 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1429 m, tlen, (long)M_TRAILINGSPACE(m)));
1434 /* Make sure we have room. */
1435 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1436 m->m_next = m_get(M_NOWAIT, MT_DATA);
1438 optp = mtod(m->m_next, u_char *);
1443 optp = (u_char *) (nth + 1);
1449 if (tp->t_flags & TF_RCVD_TSTMP) {
1450 to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1451 to.to_tsecr = tp->ts_recent;
1452 to.to_flags |= TOF_TS;
1454 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1455 /* TCP-MD5 (RFC2385). */
1456 if (tp->t_flags & TF_SIGNATURE)
1457 to.to_flags |= TOF_SIGNATURE;
1459 /* Add the options. */
1460 tlen += optlen = tcp_addoptions(&to, optp);
1462 /* Update m_len in the correct mbuf. */
1463 optm->m_len += optlen;
1469 ip6->ip6_vfc = IPV6_VERSION;
1470 ip6->ip6_nxt = IPPROTO_TCP;
1471 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1474 #if defined(INET) && defined(INET6)
1479 ip->ip_len = htons(tlen);
1480 ip->ip_ttl = V_ip_defttl;
1481 if (V_path_mtu_discovery)
1482 ip->ip_off |= htons(IP_DF);
1485 m->m_pkthdr.len = tlen;
1486 m->m_pkthdr.rcvif = NULL;
1490 * Packet is associated with a socket, so allow the
1491 * label of the response to reflect the socket label.
1493 INP_WLOCK_ASSERT(inp);
1494 mac_inpcb_create_mbuf(inp, m);
1497 * Packet is not associated with a socket, so possibly
1498 * update the label in place.
1500 mac_netinet_tcp_reply(m);
1503 nth->th_seq = htonl(seq);
1504 nth->th_ack = htonl(ack);
1506 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1507 nth->th_flags = flags;
1509 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1511 nth->th_win = htons((u_short)win);
1514 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1515 if (to.to_flags & TOF_SIGNATURE) {
1516 if (!TCPMD5_ENABLED() ||
1517 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
1524 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1527 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1528 nth->th_sum = in6_cksum_pseudo(ip6,
1529 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
1530 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
1534 #if defined(INET6) && defined(INET)
1539 m->m_pkthdr.csum_flags = CSUM_TCP;
1540 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1541 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1545 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1546 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1548 TCP_PROBE3(debug__output, tp, th, m);
1550 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
1554 TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
1555 (void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
1558 #if defined(INET) && defined(INET6)
1563 TCP_PROBE5(send, NULL, tp, ip, tp, nth);
1564 (void)ip_output(m, NULL, NULL, 0, NULL, inp);
1570 * Create a new TCP control block, making an
1571 * empty reassembly queue and hooking it to the argument
1572 * protocol control block. The `inp' parameter must have
1573 * come from the zone allocator set up in tcp_init().
1576 tcp_newtcpcb(struct inpcb *inp)
1578 struct tcpcb_mem *tm;
1581 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1584 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1589 /* Initialise cc_var struct for this tcpcb. */
1591 tp->ccv->type = IPPROTO_TCP;
1592 tp->ccv->ccvc.tcp = tp;
1593 rw_rlock(&tcp_function_lock);
1594 tp->t_fb = tcp_func_set_ptr;
1595 refcount_acquire(&tp->t_fb->tfb_refcnt);
1596 rw_runlock(&tcp_function_lock);
1598 * Use the current system default CC algorithm.
1601 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1602 CC_ALGO(tp) = CC_DEFAULT();
1605 if (CC_ALGO(tp)->cb_init != NULL)
1606 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1607 if (tp->t_fb->tfb_tcp_fb_fini)
1608 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1609 refcount_release(&tp->t_fb->tfb_refcnt);
1610 uma_zfree(V_tcpcb_zone, tm);
1616 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1617 if (tp->t_fb->tfb_tcp_fb_fini)
1618 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1619 refcount_release(&tp->t_fb->tfb_refcnt);
1620 uma_zfree(V_tcpcb_zone, tm);
1626 tp->t_vnet = inp->inp_vnet;
1628 tp->t_timers = &tm->tt;
1629 TAILQ_INIT(&tp->t_segq);
1632 isipv6 ? V_tcp_v6mssdflt :
1636 /* Set up our timeouts. */
1637 callout_init(&tp->t_timers->tt_rexmt, 1);
1638 callout_init(&tp->t_timers->tt_persist, 1);
1639 callout_init(&tp->t_timers->tt_keep, 1);
1640 callout_init(&tp->t_timers->tt_2msl, 1);
1641 callout_init(&tp->t_timers->tt_delack, 1);
1643 if (V_tcp_do_rfc1323)
1644 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1646 tp->t_flags |= TF_SACK_PERMIT;
1647 TAILQ_INIT(&tp->snd_holes);
1649 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1652 in_pcbref(inp); /* Reference for tcpcb */
1656 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1657 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
1658 * reasonable initial retransmit time.
1660 tp->t_srtt = TCPTV_SRTTBASE;
1661 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1662 tp->t_rttmin = tcp_rexmit_min;
1663 tp->t_rxtcur = TCPTV_RTOBASE;
1664 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1665 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1666 tp->t_rcvtime = ticks;
1668 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1669 * because the socket may be bound to an IPv6 wildcard address,
1670 * which may match an IPv4-mapped IPv6 address.
1672 inp->inp_ip_ttl = V_ip_defttl;
1676 * Init the TCP PCAP queues.
1678 tcp_pcap_tcpcb_init(tp);
1681 /* Initialize the per-TCPCB log data. */
1682 tcp_log_tcpcbinit(tp);
1684 if (tp->t_fb->tfb_tcp_fb_init) {
1685 (*tp->t_fb->tfb_tcp_fb_init)(tp);
1687 return (tp); /* XXX */
1691 * Switch the congestion control algorithm back to NewReno for any active
1692 * control blocks using an algorithm which is about to go away.
1693 * This ensures the CC framework can allow the unload to proceed without leaving
1694 * any dangling pointers which would trigger a panic.
1695 * Returning non-zero would inform the CC framework that something went wrong
1696 * and it would be unsafe to allow the unload to proceed. However, there is no
1697 * way for this to occur with this implementation so we always return zero.
1700 tcp_ccalgounload(struct cc_algo *unload_algo)
1702 struct cc_algo *tmpalgo;
1705 VNET_ITERATOR_DECL(vnet_iter);
1708 * Check all active control blocks across all network stacks and change
1709 * any that are using "unload_algo" back to NewReno. If "unload_algo"
1710 * requires cleanup code to be run, call it.
1713 VNET_FOREACH(vnet_iter) {
1714 CURVNET_SET(vnet_iter);
1715 INP_INFO_WLOCK(&V_tcbinfo);
1717 * New connections already part way through being initialised
1718 * with the CC algo we're removing will not race with this code
1719 * because the INP_INFO_WLOCK is held during initialisation. We
1720 * therefore don't enter the loop below until the connection
1721 * list has stabilised.
1723 CK_LIST_FOREACH(inp, &V_tcb, inp_list) {
1725 /* Important to skip tcptw structs. */
1726 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1727 (tp = intotcpcb(inp)) != NULL) {
1729 * By holding INP_WLOCK here, we are assured
1730 * that the connection is not currently
1731 * executing inside the CC module's functions
1732 * i.e. it is safe to make the switch back to
1735 if (CC_ALGO(tp) == unload_algo) {
1736 tmpalgo = CC_ALGO(tp);
1737 if (tmpalgo->cb_destroy != NULL)
1738 tmpalgo->cb_destroy(tp->ccv);
1741 * NewReno may allocate memory on
1742 * demand for certain stateful
1743 * configuration as needed, but is
1744 * coded to never fail on memory
1745 * allocation failure so it is a safe
1748 CC_ALGO(tp) = &newreno_cc_algo;
1753 INP_INFO_WUNLOCK(&V_tcbinfo);
1756 VNET_LIST_RUNLOCK();
1762 * Drop a TCP connection, reporting
1763 * the specified error. If connection is synchronized,
1764 * then send a RST to peer.
1767 tcp_drop(struct tcpcb *tp, int errno)
1769 struct socket *so = tp->t_inpcb->inp_socket;
1771 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1772 INP_WLOCK_ASSERT(tp->t_inpcb);
1774 if (TCPS_HAVERCVDSYN(tp->t_state)) {
1775 tcp_state_change(tp, TCPS_CLOSED);
1776 (void) tp->t_fb->tfb_tcp_output(tp);
1777 TCPSTAT_INC(tcps_drops);
1779 TCPSTAT_INC(tcps_conndrops);
1780 if (errno == ETIMEDOUT && tp->t_softerror)
1781 errno = tp->t_softerror;
1782 so->so_error = errno;
1783 return (tcp_close(tp));
1787 tcp_discardcb(struct tcpcb *tp)
1789 struct inpcb *inp = tp->t_inpcb;
1790 struct socket *so = inp->inp_socket;
1792 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1794 int released __unused;
1796 INP_WLOCK_ASSERT(inp);
1799 * Make sure that all of our timers are stopped before we delete the
1802 * If stopping a timer fails, we schedule a discard function in same
1803 * callout, and the last discard function called will take care of
1804 * deleting the tcpcb.
1806 tp->t_timers->tt_draincnt = 0;
1807 tcp_timer_stop(tp, TT_REXMT);
1808 tcp_timer_stop(tp, TT_PERSIST);
1809 tcp_timer_stop(tp, TT_KEEP);
1810 tcp_timer_stop(tp, TT_2MSL);
1811 tcp_timer_stop(tp, TT_DELACK);
1812 if (tp->t_fb->tfb_tcp_timer_stop_all) {
1814 * Call the stop-all function of the methods,
1815 * this function should call the tcp_timer_stop()
1816 * method with each of the function specific timeouts.
1817 * That stop will be called via the tfb_tcp_timer_stop()
1818 * which should use the async drain function of the
1819 * callout system (see tcp_var.h).
1821 tp->t_fb->tfb_tcp_timer_stop_all(tp);
1825 * If we got enough samples through the srtt filter,
1826 * save the rtt and rttvar in the routing entry.
1827 * 'Enough' is arbitrarily defined as 4 rtt samples.
1828 * 4 samples is enough for the srtt filter to converge
1829 * to within enough % of the correct value; fewer samples
1830 * and we could save a bogus rtt. The danger is not high
1831 * as tcp quickly recovers from everything.
1832 * XXX: Works very well but needs some more statistics!
1834 if (tp->t_rttupdated >= 4) {
1835 struct hc_metrics_lite metrics;
1838 bzero(&metrics, sizeof(metrics));
1840 * Update the ssthresh always when the conditions below
1841 * are satisfied. This gives us better new start value
1842 * for the congestion avoidance for new connections.
1843 * ssthresh is only set if packet loss occurred on a session.
1845 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1846 * being torn down. Ideally this code would not use 'so'.
1848 ssthresh = tp->snd_ssthresh;
1849 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1851 * convert the limit from user data bytes to
1852 * packets then to packet data bytes.
1854 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1857 ssthresh *= (tp->t_maxseg +
1859 (isipv6 ? sizeof (struct ip6_hdr) +
1860 sizeof (struct tcphdr) :
1862 sizeof (struct tcpiphdr)
1869 metrics.rmx_ssthresh = ssthresh;
1871 metrics.rmx_rtt = tp->t_srtt;
1872 metrics.rmx_rttvar = tp->t_rttvar;
1873 metrics.rmx_cwnd = tp->snd_cwnd;
1874 metrics.rmx_sendpipe = 0;
1875 metrics.rmx_recvpipe = 0;
1877 tcp_hc_update(&inp->inp_inc, &metrics);
1880 /* free the reassembly queue, if any */
1881 tcp_reass_flush(tp);
1884 /* Disconnect offload device, if any. */
1885 if (tp->t_flags & TF_TOE)
1886 tcp_offload_detach(tp);
1889 tcp_free_sackholes(tp);
1892 /* Free the TCP PCAP queues. */
1893 tcp_pcap_drain(&(tp->t_inpkts));
1894 tcp_pcap_drain(&(tp->t_outpkts));
1897 /* Allow the CC algorithm to clean up after itself. */
1898 if (CC_ALGO(tp)->cb_destroy != NULL)
1899 CC_ALGO(tp)->cb_destroy(tp->ccv);
1903 khelp_destroy_osd(tp->osd);
1907 inp->inp_ppcb = NULL;
1908 if (tp->t_timers->tt_draincnt == 0) {
1909 /* We own the last reference on tcpcb, let's free it. */
1911 tcp_log_tcpcbfini(tp);
1913 TCPSTATES_DEC(tp->t_state);
1914 if (tp->t_fb->tfb_tcp_fb_fini)
1915 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1916 refcount_release(&tp->t_fb->tfb_refcnt);
1918 uma_zfree(V_tcpcb_zone, tp);
1919 released = in_pcbrele_wlocked(inp);
1920 KASSERT(!released, ("%s: inp %p should not have been released "
1921 "here", __func__, inp));
1926 tcp_timer_discard(void *ptp)
1930 struct epoch_tracker et;
1932 tp = (struct tcpcb *)ptp;
1933 CURVNET_SET(tp->t_vnet);
1934 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
1936 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1939 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1940 ("%s: tcpcb has to be stopped here", __func__));
1941 tp->t_timers->tt_draincnt--;
1942 if (tp->t_timers->tt_draincnt == 0) {
1943 /* We own the last reference on this tcpcb, let's free it. */
1945 tcp_log_tcpcbfini(tp);
1947 TCPSTATES_DEC(tp->t_state);
1948 if (tp->t_fb->tfb_tcp_fb_fini)
1949 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1950 refcount_release(&tp->t_fb->tfb_refcnt);
1952 uma_zfree(V_tcpcb_zone, tp);
1953 if (in_pcbrele_wlocked(inp)) {
1954 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1960 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1965 * Attempt to close a TCP control block, marking it as dropped, and freeing
1966 * the socket if we hold the only reference.
1969 tcp_close(struct tcpcb *tp)
1971 struct inpcb *inp = tp->t_inpcb;
1974 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1975 INP_WLOCK_ASSERT(inp);
1978 if (tp->t_state == TCPS_LISTEN)
1979 tcp_offload_listen_stop(tp);
1982 * This releases the TFO pending counter resource for TFO listen
1983 * sockets as well as passively-created TFO sockets that transition
1984 * from SYN_RECEIVED to CLOSED.
1986 if (tp->t_tfo_pending) {
1987 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
1988 tp->t_tfo_pending = NULL;
1991 TCPSTAT_INC(tcps_closed);
1992 if (tp->t_state != TCPS_CLOSED)
1993 tcp_state_change(tp, TCPS_CLOSED);
1994 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1995 so = inp->inp_socket;
1996 soisdisconnected(so);
1997 if (inp->inp_flags & INP_SOCKREF) {
1998 KASSERT(so->so_state & SS_PROTOREF,
1999 ("tcp_close: !SS_PROTOREF"));
2000 inp->inp_flags &= ~INP_SOCKREF;
2003 so->so_state &= ~SS_PROTOREF;
2013 VNET_ITERATOR_DECL(vnet_iter);
2018 VNET_LIST_RLOCK_NOSLEEP();
2019 VNET_FOREACH(vnet_iter) {
2020 CURVNET_SET(vnet_iter);
2025 * Walk the tcpbs, if existing, and flush the reassembly queue,
2026 * if there is one...
2027 * XXX: The "Net/3" implementation doesn't imply that the TCP
2028 * reassembly queue should be flushed, but in a situation
2029 * where we're really low on mbufs, this is potentially
2032 INP_INFO_WLOCK(&V_tcbinfo);
2033 CK_LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
2035 if (inpb->inp_flags & INP_TIMEWAIT) {
2039 if ((tcpb = intotcpcb(inpb)) != NULL) {
2040 tcp_reass_flush(tcpb);
2041 tcp_clean_sackreport(tcpb);
2043 tcp_log_drain(tcpb);
2046 if (tcp_pcap_aggressive_free) {
2047 /* Free the TCP PCAP queues. */
2048 tcp_pcap_drain(&(tcpb->t_inpkts));
2049 tcp_pcap_drain(&(tcpb->t_outpkts));
2055 INP_INFO_WUNLOCK(&V_tcbinfo);
2058 VNET_LIST_RUNLOCK_NOSLEEP();
2062 * Notify a tcp user of an asynchronous error;
2063 * store error as soft error, but wake up user
2064 * (for now, won't do anything until can select for soft error).
2066 * Do not wake up user since there currently is no mechanism for
2067 * reporting soft errors (yet - a kqueue filter may be added).
2069 static struct inpcb *
2070 tcp_notify(struct inpcb *inp, int error)
2074 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2075 INP_WLOCK_ASSERT(inp);
2077 if ((inp->inp_flags & INP_TIMEWAIT) ||
2078 (inp->inp_flags & INP_DROPPED))
2081 tp = intotcpcb(inp);
2082 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
2085 * Ignore some errors if we are hooked up.
2086 * If connection hasn't completed, has retransmitted several times,
2087 * and receives a second error, give up now. This is better
2088 * than waiting a long time to establish a connection that
2089 * can never complete.
2091 if (tp->t_state == TCPS_ESTABLISHED &&
2092 (error == EHOSTUNREACH || error == ENETUNREACH ||
2093 error == EHOSTDOWN)) {
2094 if (inp->inp_route.ro_rt) {
2095 RTFREE(inp->inp_route.ro_rt);
2096 inp->inp_route.ro_rt = (struct rtentry *)NULL;
2099 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
2101 tp = tcp_drop(tp, error);
2107 tp->t_softerror = error;
2111 wakeup( &so->so_timeo);
2118 tcp_pcblist(SYSCTL_HANDLER_ARGS)
2120 int error, i, m, n, pcb_count;
2121 struct inpcb *inp, **inp_list;
2124 struct epoch_tracker et;
2127 * The process of preparing the TCB list is too time-consuming and
2128 * resource-intensive to repeat twice on every request.
2130 if (req->oldptr == NULL) {
2131 n = V_tcbinfo.ipi_count +
2132 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2133 n += imax(n / 8, 10);
2134 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
2138 if (req->newptr != NULL)
2142 * OK, now we're committed to doing something.
2144 INP_LIST_RLOCK(&V_tcbinfo);
2145 gencnt = V_tcbinfo.ipi_gencnt;
2146 n = V_tcbinfo.ipi_count;
2147 INP_LIST_RUNLOCK(&V_tcbinfo);
2149 m = counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2151 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
2152 + (n + m) * sizeof(struct xtcpcb));
2156 xig.xig_len = sizeof xig;
2157 xig.xig_count = n + m;
2158 xig.xig_gen = gencnt;
2159 xig.xig_sogen = so_gencnt;
2160 error = SYSCTL_OUT(req, &xig, sizeof xig);
2164 error = syncache_pcblist(req, m, &pcb_count);
2168 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
2170 INP_INFO_WLOCK(&V_tcbinfo);
2171 for (inp = CK_LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
2172 inp != NULL && i < n; inp = CK_LIST_NEXT(inp, inp_list)) {
2174 if (inp->inp_gencnt <= gencnt) {
2176 * XXX: This use of cr_cansee(), introduced with
2177 * TCP state changes, is not quite right, but for
2178 * now, better than nothing.
2180 if (inp->inp_flags & INP_TIMEWAIT) {
2181 if (intotw(inp) != NULL)
2182 error = cr_cansee(req->td->td_ucred,
2183 intotw(inp)->tw_cred);
2185 error = EINVAL; /* Skip this inp. */
2187 error = cr_canseeinpcb(req->td->td_ucred, inp);
2190 inp_list[i++] = inp;
2195 INP_INFO_WUNLOCK(&V_tcbinfo);
2199 for (i = 0; i < n; i++) {
2202 if (inp->inp_gencnt <= gencnt) {
2205 tcp_inptoxtp(inp, &xt);
2207 error = SYSCTL_OUT(req, &xt, sizeof xt);
2211 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
2212 for (i = 0; i < n; i++) {
2215 if (!in_pcbrele_rlocked(inp))
2218 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2222 * Give the user an updated idea of our state.
2223 * If the generation differs from what we told
2224 * her before, she knows that something happened
2225 * while we were processing this request, and it
2226 * might be necessary to retry.
2228 INP_LIST_RLOCK(&V_tcbinfo);
2229 xig.xig_gen = V_tcbinfo.ipi_gencnt;
2230 xig.xig_sogen = so_gencnt;
2231 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
2232 INP_LIST_RUNLOCK(&V_tcbinfo);
2233 error = SYSCTL_OUT(req, &xig, sizeof xig);
2235 free(inp_list, M_TEMP);
2239 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
2240 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
2241 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
2245 tcp_getcred(SYSCTL_HANDLER_ARGS)
2248 struct sockaddr_in addrs[2];
2252 error = priv_check(req->td, PRIV_NETINET_GETCRED);
2255 error = SYSCTL_IN(req, addrs, sizeof(addrs));
2258 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
2259 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
2261 if (inp->inp_socket == NULL)
2264 error = cr_canseeinpcb(req->td->td_ucred, inp);
2266 cru2x(inp->inp_cred, &xuc);
2271 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2275 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
2276 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
2277 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
2282 tcp6_getcred(SYSCTL_HANDLER_ARGS)
2285 struct sockaddr_in6 addrs[2];
2292 error = priv_check(req->td, PRIV_NETINET_GETCRED);
2295 error = SYSCTL_IN(req, addrs, sizeof(addrs));
2298 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
2299 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
2302 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
2304 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
2313 inp = in_pcblookup(&V_tcbinfo,
2314 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
2316 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
2317 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
2320 inp = in6_pcblookup(&V_tcbinfo,
2321 &addrs[1].sin6_addr, addrs[1].sin6_port,
2322 &addrs[0].sin6_addr, addrs[0].sin6_port,
2323 INPLOOKUP_RLOCKPCB, NULL);
2325 if (inp->inp_socket == NULL)
2328 error = cr_canseeinpcb(req->td->td_ucred, inp);
2330 cru2x(inp->inp_cred, &xuc);
2335 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2339 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
2340 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
2341 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
2347 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
2349 struct ip *ip = vip;
2351 struct in_addr faddr;
2354 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2356 struct in_conninfo inc;
2357 struct epoch_tracker et;
2358 tcp_seq icmp_tcp_seq;
2361 faddr = ((struct sockaddr_in *)sa)->sin_addr;
2362 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
2365 if (cmd == PRC_MSGSIZE)
2366 notify = tcp_mtudisc_notify;
2367 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2368 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2369 cmd == PRC_TIMXCEED_INTRANS) && ip)
2370 notify = tcp_drop_syn_sent;
2373 * Hostdead is ugly because it goes linearly through all PCBs.
2374 * XXX: We never get this from ICMP, otherwise it makes an
2375 * excellent DoS attack on machines with many connections.
2377 else if (cmd == PRC_HOSTDEAD)
2379 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
2383 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
2387 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
2388 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2389 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
2390 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
2391 th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2392 if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2393 /* signal EHOSTDOWN, as it flushes the cached route */
2394 inp = (*notify)(inp, EHOSTDOWN);
2397 icmp_tcp_seq = th->th_seq;
2399 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2400 !(inp->inp_flags & INP_DROPPED) &&
2401 !(inp->inp_socket == NULL)) {
2402 tp = intotcpcb(inp);
2403 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2404 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2405 if (cmd == PRC_MSGSIZE) {
2408 * If we got a needfrag set the MTU
2409 * in the route to the suggested new
2410 * value (if given) and then notify.
2412 mtu = ntohs(icp->icmp_nextmtu);
2414 * If no alternative MTU was
2415 * proposed, try the next smaller
2420 ntohs(ip->ip_len), 1);
2421 if (mtu < V_tcp_minmss +
2422 sizeof(struct tcpiphdr))
2423 mtu = V_tcp_minmss +
2424 sizeof(struct tcpiphdr);
2426 * Only process the offered MTU if it
2427 * is smaller than the current one.
2429 if (mtu < tp->t_maxseg +
2430 sizeof(struct tcpiphdr)) {
2431 bzero(&inc, sizeof(inc));
2432 inc.inc_faddr = faddr;
2434 inp->inp_inc.inc_fibnum;
2435 tcp_hc_updatemtu(&inc, mtu);
2436 tcp_mtudisc(inp, mtu);
2439 inp = (*notify)(inp,
2440 inetctlerrmap[cmd]);
2444 bzero(&inc, sizeof(inc));
2445 inc.inc_fport = th->th_dport;
2446 inc.inc_lport = th->th_sport;
2447 inc.inc_faddr = faddr;
2448 inc.inc_laddr = ip->ip_src;
2449 syncache_unreach(&inc, icmp_tcp_seq);
2454 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2460 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
2462 struct in6_addr *dst;
2463 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2464 struct ip6_hdr *ip6;
2468 struct icmp6_hdr *icmp6;
2469 struct ip6ctlparam *ip6cp = NULL;
2470 const struct sockaddr_in6 *sa6_src = NULL;
2471 struct in_conninfo inc;
2472 struct epoch_tracker et;
2477 tcp_seq icmp_tcp_seq;
2481 if (sa->sa_family != AF_INET6 ||
2482 sa->sa_len != sizeof(struct sockaddr_in6))
2485 /* if the parameter is from icmp6, decode it. */
2487 ip6cp = (struct ip6ctlparam *)d;
2488 icmp6 = ip6cp->ip6c_icmp6;
2490 ip6 = ip6cp->ip6c_ip6;
2491 off = ip6cp->ip6c_off;
2492 sa6_src = ip6cp->ip6c_src;
2493 dst = ip6cp->ip6c_finaldst;
2497 off = 0; /* fool gcc */
2502 if (cmd == PRC_MSGSIZE)
2503 notify = tcp_mtudisc_notify;
2504 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2505 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2506 cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
2507 notify = tcp_drop_syn_sent;
2510 * Hostdead is ugly because it goes linearly through all PCBs.
2511 * XXX: We never get this from ICMP, otherwise it makes an
2512 * excellent DoS attack on machines with many connections.
2514 else if (cmd == PRC_HOSTDEAD)
2516 else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
2520 in6_pcbnotify(&V_tcbinfo, sa, 0,
2521 (const struct sockaddr *)sa6_src,
2522 0, cmd, NULL, notify);
2526 /* Check if we can safely get the ports from the tcp hdr */
2529 (int32_t) (off + sizeof(struct tcp_ports)))) {
2532 bzero(&t_ports, sizeof(struct tcp_ports));
2533 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
2534 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
2535 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
2536 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
2537 if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2538 /* signal EHOSTDOWN, as it flushes the cached route */
2539 inp = (*notify)(inp, EHOSTDOWN);
2542 off += sizeof(struct tcp_ports);
2543 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
2546 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
2548 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2549 !(inp->inp_flags & INP_DROPPED) &&
2550 !(inp->inp_socket == NULL)) {
2551 tp = intotcpcb(inp);
2552 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2553 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2554 if (cmd == PRC_MSGSIZE) {
2557 * If we got a needfrag set the MTU
2558 * in the route to the suggested new
2559 * value (if given) and then notify.
2561 mtu = ntohl(icmp6->icmp6_mtu);
2563 * If no alternative MTU was
2564 * proposed, or the proposed
2565 * MTU was too small, set to
2568 if (mtu < IPV6_MMTU)
2569 mtu = IPV6_MMTU - 8;
2570 bzero(&inc, sizeof(inc));
2571 inc.inc_fibnum = M_GETFIB(m);
2572 inc.inc_flags |= INC_ISIPV6;
2573 inc.inc6_faddr = *dst;
2574 if (in6_setscope(&inc.inc6_faddr,
2575 m->m_pkthdr.rcvif, NULL))
2578 * Only process the offered MTU if it
2579 * is smaller than the current one.
2581 if (mtu < tp->t_maxseg +
2582 sizeof (struct tcphdr) +
2583 sizeof (struct ip6_hdr)) {
2584 tcp_hc_updatemtu(&inc, mtu);
2585 tcp_mtudisc(inp, mtu);
2586 ICMP6STAT_INC(icp6s_pmtuchg);
2589 inp = (*notify)(inp,
2590 inet6ctlerrmap[cmd]);
2594 bzero(&inc, sizeof(inc));
2595 inc.inc_fibnum = M_GETFIB(m);
2596 inc.inc_flags |= INC_ISIPV6;
2597 inc.inc_fport = t_ports.th_dport;
2598 inc.inc_lport = t_ports.th_sport;
2599 inc.inc6_faddr = *dst;
2600 inc.inc6_laddr = ip6->ip6_src;
2601 syncache_unreach(&inc, icmp_tcp_seq);
2606 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2611 tcp_keyed_hash(struct in_conninfo *inc, u_char *key)
2617 MD5Update(&ctx, &inc->inc_fport, sizeof(uint16_t));
2618 MD5Update(&ctx, &inc->inc_lport, sizeof(uint16_t));
2619 switch (inc->inc_flags & INC_ISIPV6) {
2622 MD5Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr));
2623 MD5Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr));
2628 MD5Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr));
2629 MD5Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr));
2633 MD5Update(&ctx, key, 32);
2634 MD5Final((unsigned char *)hash, &ctx);
2640 tcp_new_ts_offset(struct in_conninfo *inc)
2642 return (tcp_keyed_hash(inc, V_ts_offset_secret));
2646 * Following is where TCP initial sequence number generation occurs.
2648 * There are two places where we must use initial sequence numbers:
2649 * 1. In SYN-ACK packets.
2650 * 2. In SYN packets.
2652 * All ISNs for SYN-ACK packets are generated by the syncache. See
2653 * tcp_syncache.c for details.
2655 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2656 * depends on this property. In addition, these ISNs should be
2657 * unguessable so as to prevent connection hijacking. To satisfy
2658 * the requirements of this situation, the algorithm outlined in
2659 * RFC 1948 is used, with only small modifications.
2661 * Implementation details:
2663 * Time is based off the system timer, and is corrected so that it
2664 * increases by one megabyte per second. This allows for proper
2665 * recycling on high speed LANs while still leaving over an hour
2668 * As reading the *exact* system time is too expensive to be done
2669 * whenever setting up a TCP connection, we increment the time
2670 * offset in two ways. First, a small random positive increment
2671 * is added to isn_offset for each connection that is set up.
2672 * Second, the function tcp_isn_tick fires once per clock tick
2673 * and increments isn_offset as necessary so that sequence numbers
2674 * are incremented at approximately ISN_BYTES_PER_SECOND. The
2675 * random positive increments serve only to ensure that the same
2676 * exact sequence number is never sent out twice (as could otherwise
2677 * happen when a port is recycled in less than the system tick
2680 * net.inet.tcp.isn_reseed_interval controls the number of seconds
2681 * between seeding of isn_secret. This is normally set to zero,
2682 * as reseeding should not be necessary.
2684 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2685 * isn_offset_old, and isn_ctx is performed using the ISN lock. In
2686 * general, this means holding an exclusive (write) lock.
2689 #define ISN_BYTES_PER_SECOND 1048576
2690 #define ISN_STATIC_INCREMENT 4096
2691 #define ISN_RANDOM_INCREMENT (4096 - 1)
2693 VNET_DEFINE_STATIC(u_char, isn_secret[32]);
2694 VNET_DEFINE_STATIC(int, isn_last);
2695 VNET_DEFINE_STATIC(int, isn_last_reseed);
2696 VNET_DEFINE_STATIC(u_int32_t, isn_offset);
2697 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old);
2699 #define V_isn_secret VNET(isn_secret)
2700 #define V_isn_last VNET(isn_last)
2701 #define V_isn_last_reseed VNET(isn_last_reseed)
2702 #define V_isn_offset VNET(isn_offset)
2703 #define V_isn_offset_old VNET(isn_offset_old)
2706 tcp_new_isn(struct in_conninfo *inc)
2709 u_int32_t projected_offset;
2712 /* Seed if this is the first use, reseed if requested. */
2713 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2714 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2716 read_random(&V_isn_secret, sizeof(V_isn_secret));
2717 V_isn_last_reseed = ticks;
2720 /* Compute the md5 hash and return the ISN. */
2721 new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret);
2722 V_isn_offset += ISN_STATIC_INCREMENT +
2723 (arc4random() & ISN_RANDOM_INCREMENT);
2724 if (ticks != V_isn_last) {
2725 projected_offset = V_isn_offset_old +
2726 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2727 if (SEQ_GT(projected_offset, V_isn_offset))
2728 V_isn_offset = projected_offset;
2729 V_isn_offset_old = V_isn_offset;
2732 new_isn += V_isn_offset;
2738 * When a specific ICMP unreachable message is received and the
2739 * connection state is SYN-SENT, drop the connection. This behavior
2740 * is controlled by the icmp_may_rst sysctl.
2743 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2747 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2748 INP_WLOCK_ASSERT(inp);
2750 if ((inp->inp_flags & INP_TIMEWAIT) ||
2751 (inp->inp_flags & INP_DROPPED))
2754 tp = intotcpcb(inp);
2755 if (tp->t_state != TCPS_SYN_SENT)
2758 if (IS_FASTOPEN(tp->t_flags))
2759 tcp_fastopen_disable_path(tp);
2761 tp = tcp_drop(tp, errno);
2769 * When `need fragmentation' ICMP is received, update our idea of the MSS
2770 * based on the new value. Also nudge TCP to send something, since we
2771 * know the packet we just sent was dropped.
2772 * This duplicates some code in the tcp_mss() function in tcp_input.c.
2774 static struct inpcb *
2775 tcp_mtudisc_notify(struct inpcb *inp, int error)
2778 tcp_mtudisc(inp, -1);
2783 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2788 INP_WLOCK_ASSERT(inp);
2789 if ((inp->inp_flags & INP_TIMEWAIT) ||
2790 (inp->inp_flags & INP_DROPPED))
2793 tp = intotcpcb(inp);
2794 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2796 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2798 so = inp->inp_socket;
2799 SOCKBUF_LOCK(&so->so_snd);
2800 /* If the mss is larger than the socket buffer, decrease the mss. */
2801 if (so->so_snd.sb_hiwat < tp->t_maxseg)
2802 tp->t_maxseg = so->so_snd.sb_hiwat;
2803 SOCKBUF_UNLOCK(&so->so_snd);
2805 TCPSTAT_INC(tcps_mturesent);
2807 tp->snd_nxt = tp->snd_una;
2808 tcp_free_sackholes(tp);
2809 tp->snd_recover = tp->snd_max;
2810 if (tp->t_flags & TF_SACK_PERMIT)
2811 EXIT_FASTRECOVERY(tp->t_flags);
2812 tp->t_fb->tfb_tcp_output(tp);
2817 * Look-up the routing entry to the peer of this inpcb. If no route
2818 * is found and it cannot be allocated, then return 0. This routine
2819 * is called by TCP routines that access the rmx structure and by
2820 * tcp_mss_update to get the peer/interface MTU.
2823 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2825 struct nhop4_extended nh4;
2827 uint32_t maxmtu = 0;
2829 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2831 if (inc->inc_faddr.s_addr != INADDR_ANY) {
2833 if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr,
2834 NHR_REF, 0, &nh4) != 0)
2838 maxmtu = nh4.nh_mtu;
2840 /* Report additional interface capabilities. */
2842 if (ifp->if_capenable & IFCAP_TSO4 &&
2843 ifp->if_hwassist & CSUM_TSO) {
2844 cap->ifcap |= CSUM_TSO;
2845 cap->tsomax = ifp->if_hw_tsomax;
2846 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2847 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2850 fib4_free_nh_ext(inc->inc_fibnum, &nh4);
2858 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2860 struct nhop6_extended nh6;
2861 struct in6_addr dst6;
2864 uint32_t maxmtu = 0;
2866 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2868 if (inc->inc_flags & INC_IPV6MINMTU)
2871 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2872 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
2873 if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0,
2878 maxmtu = nh6.nh_mtu;
2880 /* Report additional interface capabilities. */
2882 if (ifp->if_capenable & IFCAP_TSO6 &&
2883 ifp->if_hwassist & CSUM_TSO) {
2884 cap->ifcap |= CSUM_TSO;
2885 cap->tsomax = ifp->if_hw_tsomax;
2886 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2887 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2890 fib6_free_nh_ext(inc->inc_fibnum, &nh6);
2898 * Calculate effective SMSS per RFC5681 definition for a given TCP
2899 * connection at its current state, taking into account SACK and etc.
2902 tcp_maxseg(const struct tcpcb *tp)
2906 if (tp->t_flags & TF_NOOPT)
2907 return (tp->t_maxseg);
2910 * Here we have a simplified code from tcp_addoptions(),
2911 * without a proper loop, and having most of paddings hardcoded.
2912 * We might make mistakes with padding here in some edge cases,
2913 * but this is harmless, since result of tcp_maxseg() is used
2914 * only in cwnd and ssthresh estimations.
2916 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4)
2917 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2918 if (tp->t_flags & TF_RCVD_TSTMP)
2919 optlen = TCPOLEN_TSTAMP_APPA;
2922 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2923 if (tp->t_flags & TF_SIGNATURE)
2924 optlen += PAD(TCPOLEN_SIGNATURE);
2926 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
2927 optlen += TCPOLEN_SACKHDR;
2928 optlen += tp->rcv_numsacks * TCPOLEN_SACK;
2929 optlen = PAD(optlen);
2932 if (tp->t_flags & TF_REQ_TSTMP)
2933 optlen = TCPOLEN_TSTAMP_APPA;
2935 optlen = PAD(TCPOLEN_MAXSEG);
2936 if (tp->t_flags & TF_REQ_SCALE)
2937 optlen += PAD(TCPOLEN_WINDOW);
2938 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2939 if (tp->t_flags & TF_SIGNATURE)
2940 optlen += PAD(TCPOLEN_SIGNATURE);
2942 if (tp->t_flags & TF_SACK_PERMIT)
2943 optlen += PAD(TCPOLEN_SACK_PERMITTED);
2946 optlen = min(optlen, TCP_MAXOLEN);
2947 return (tp->t_maxseg - optlen);
2951 sysctl_drop(SYSCTL_HANDLER_ARGS)
2953 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2954 struct sockaddr_storage addrs[2];
2958 struct sockaddr_in *fin, *lin;
2959 struct epoch_tracker et;
2961 struct sockaddr_in6 *fin6, *lin6;
2972 if (req->oldptr != NULL || req->oldlen != 0)
2974 if (req->newptr == NULL)
2976 if (req->newlen < sizeof(addrs))
2978 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2982 switch (addrs[0].ss_family) {
2985 fin6 = (struct sockaddr_in6 *)&addrs[0];
2986 lin6 = (struct sockaddr_in6 *)&addrs[1];
2987 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2988 lin6->sin6_len != sizeof(struct sockaddr_in6))
2990 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2991 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2993 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2994 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2995 fin = (struct sockaddr_in *)&addrs[0];
2996 lin = (struct sockaddr_in *)&addrs[1];
2999 error = sa6_embedscope(fin6, V_ip6_use_defzone);
3002 error = sa6_embedscope(lin6, V_ip6_use_defzone);
3009 fin = (struct sockaddr_in *)&addrs[0];
3010 lin = (struct sockaddr_in *)&addrs[1];
3011 if (fin->sin_len != sizeof(struct sockaddr_in) ||
3012 lin->sin_len != sizeof(struct sockaddr_in))
3019 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
3020 switch (addrs[0].ss_family) {
3023 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3024 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3025 INPLOOKUP_WLOCKPCB, NULL);
3030 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3031 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3036 if (inp->inp_flags & INP_TIMEWAIT) {
3038 * XXXRW: There currently exists a state where an
3039 * inpcb is present, but its timewait state has been
3040 * discarded. For now, don't allow dropping of this
3048 } else if (!(inp->inp_flags & INP_DROPPED) &&
3049 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
3050 tp = intotcpcb(inp);
3051 tp = tcp_drop(tp, ECONNABORTED);
3058 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
3062 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
3063 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
3064 0, sysctl_drop, "", "Drop TCP connection");
3067 * Generate a standardized TCP log line for use throughout the
3068 * tcp subsystem. Memory allocation is done with M_NOWAIT to
3069 * allow use in the interrupt context.
3071 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
3072 * NB: The function may return NULL if memory allocation failed.
3074 * Due to header inclusion and ordering limitations the struct ip
3075 * and ip6_hdr pointers have to be passed as void pointers.
3078 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3082 /* Is logging enabled? */
3083 if (tcp_log_in_vain == 0)
3086 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3090 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3094 /* Is logging enabled? */
3095 if (tcp_log_debug == 0)
3098 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3102 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3109 const struct ip6_hdr *ip6;
3111 ip6 = (const struct ip6_hdr *)ip6hdr;
3113 ip = (struct ip *)ip4hdr;
3116 * The log line looks like this:
3117 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
3119 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
3120 sizeof(PRINT_TH_FLAGS) + 1 +
3122 2 * INET6_ADDRSTRLEN;
3124 2 * INET_ADDRSTRLEN;
3127 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
3131 strcat(s, "TCP: [");
3134 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
3135 inet_ntoa_r(inc->inc_faddr, sp);
3137 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3139 inet_ntoa_r(inc->inc_laddr, sp);
3141 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3144 ip6_sprintf(sp, &inc->inc6_faddr);
3146 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3148 ip6_sprintf(sp, &inc->inc6_laddr);
3150 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3151 } else if (ip6 && th) {
3152 ip6_sprintf(sp, &ip6->ip6_src);
3154 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3156 ip6_sprintf(sp, &ip6->ip6_dst);
3158 sprintf(sp, "]:%i", ntohs(th->th_dport));
3161 } else if (ip && th) {
3162 inet_ntoa_r(ip->ip_src, sp);
3164 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3166 inet_ntoa_r(ip->ip_dst, sp);
3168 sprintf(sp, "]:%i", ntohs(th->th_dport));
3176 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
3177 if (*(s + size - 1) != '\0')
3178 panic("%s: string too long", __func__);
3183 * A subroutine which makes it easy to track TCP state changes with DTrace.
3184 * This function shouldn't be called for t_state initializations that don't
3185 * correspond to actual TCP state transitions.
3188 tcp_state_change(struct tcpcb *tp, int newstate)
3190 #if defined(KDTRACE_HOOKS)
3191 int pstate = tp->t_state;
3194 TCPSTATES_DEC(tp->t_state);
3195 TCPSTATES_INC(newstate);
3196 tp->t_state = newstate;
3197 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
3201 * Create an external-format (``xtcpcb'') structure using the information in
3202 * the kernel-format tcpcb structure pointed to by tp. This is done to
3203 * reduce the spew of irrelevant information over this interface, to isolate
3204 * user code from changes in the kernel structure, and potentially to provide
3205 * information-hiding if we decide that some of this information should be
3206 * hidden from users.
3209 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
3211 struct tcpcb *tp = intotcpcb(inp);
3214 if (inp->inp_flags & INP_TIMEWAIT) {
3215 bzero(xt, sizeof(struct xtcpcb));
3216 xt->t_state = TCPS_TIME_WAIT;
3218 xt->t_state = tp->t_state;
3219 xt->t_logstate = tp->t_logstate;
3220 xt->t_flags = tp->t_flags;
3221 xt->t_sndzerowin = tp->t_sndzerowin;
3222 xt->t_sndrexmitpack = tp->t_sndrexmitpack;
3223 xt->t_rcvoopack = tp->t_rcvoopack;
3225 now = getsbinuptime();
3226 #define COPYTIMER(ttt) do { \
3227 if (callout_active(&tp->t_timers->ttt)) \
3228 xt->ttt = (tp->t_timers->ttt.c_time - now) / \
3233 COPYTIMER(tt_delack);
3234 COPYTIMER(tt_rexmt);
3235 COPYTIMER(tt_persist);
3239 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
3241 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
3242 TCP_FUNCTION_NAME_LEN_MAX);
3243 bzero(xt->xt_logid, TCP_LOG_ID_LEN);
3245 (void)tcp_log_get_id(tp, xt->xt_logid);
3249 xt->xt_len = sizeof(struct xtcpcb);
3250 in_pcbtoxinpcb(inp, &xt->xt_inp);
3251 if (inp->inp_socket == NULL)
3252 xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP;