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_kern_tls.h"
41 #include "opt_tcpdebug.h"
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/callout.h>
46 #include <sys/eventhandler.h>
48 #include <sys/hhook.h>
50 #include <sys/kernel.h>
52 #include <sys/khelp.h>
57 #include <sys/sysctl.h>
59 #include <sys/malloc.h>
60 #include <sys/refcount.h>
63 #include <sys/domain.h>
68 #include <sys/socket.h>
69 #include <sys/socketvar.h>
70 #include <sys/protosw.h>
71 #include <sys/random.h>
75 #include <net/route.h>
77 #include <net/if_var.h>
80 #include <netinet/in.h>
81 #include <netinet/in_fib.h>
82 #include <netinet/in_kdtrace.h>
83 #include <netinet/in_pcb.h>
84 #include <netinet/in_systm.h>
85 #include <netinet/in_var.h>
86 #include <netinet/ip.h>
87 #include <netinet/ip_icmp.h>
88 #include <netinet/ip_var.h>
90 #include <netinet/icmp6.h>
91 #include <netinet/ip6.h>
92 #include <netinet6/in6_fib.h>
93 #include <netinet6/in6_pcb.h>
94 #include <netinet6/ip6_var.h>
95 #include <netinet6/scope6_var.h>
96 #include <netinet6/nd6.h>
99 #include <netinet/tcp.h>
100 #include <netinet/tcp_fsm.h>
101 #include <netinet/tcp_seq.h>
102 #include <netinet/tcp_timer.h>
103 #include <netinet/tcp_var.h>
104 #include <netinet/tcp_log_buf.h>
105 #include <netinet/tcp_syncache.h>
106 #include <netinet/tcp_hpts.h>
107 #include <netinet/cc/cc.h>
109 #include <netinet6/tcp6_var.h>
111 #include <netinet/tcpip.h>
112 #include <netinet/tcp_fastopen.h>
114 #include <netinet/tcp_pcap.h>
117 #include <netinet/tcp_debug.h>
120 #include <netinet6/ip6protosw.h>
123 #include <netinet/tcp_offload.h>
126 #include <netipsec/ipsec_support.h>
128 #include <machine/in_cksum.h>
131 #include <security/mac/mac_framework.h>
133 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
135 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
138 struct rwlock tcp_function_lock;
141 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
146 error = sysctl_handle_int(oidp, &new, 0, req);
147 if (error == 0 && req->newptr) {
148 if (new < TCP_MINMSS)
156 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
157 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
158 &sysctl_net_inet_tcp_mss_check, "I",
159 "Default TCP Maximum Segment Size");
163 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
167 new = V_tcp_v6mssdflt;
168 error = sysctl_handle_int(oidp, &new, 0, req);
169 if (error == 0 && req->newptr) {
170 if (new < TCP_MINMSS)
173 V_tcp_v6mssdflt = new;
178 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
179 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
180 &sysctl_net_inet_tcp_mss_v6_check, "I",
181 "Default TCP Maximum Segment Size for IPv6");
185 * Minimum MSS we accept and use. This prevents DoS attacks where
186 * we are forced to a ridiculous low MSS like 20 and send hundreds
187 * of packets instead of one. The effect scales with the available
188 * bandwidth and quickly saturates the CPU and network interface
189 * with packet generation and sending. Set to zero to disable MINMSS
190 * checking. This setting prevents us from sending too small packets.
192 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
193 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
194 &VNET_NAME(tcp_minmss), 0,
195 "Minimum TCP Maximum Segment Size");
197 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
198 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
199 &VNET_NAME(tcp_do_rfc1323), 0,
200 "Enable rfc1323 (high performance TCP) extensions");
202 VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1;
203 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW,
204 &VNET_NAME(tcp_ts_offset_per_conn), 0,
205 "Initialize TCP timestamps per connection instead of per host pair");
207 static int tcp_log_debug = 0;
208 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
209 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
211 static int tcp_tcbhashsize;
212 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
213 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
215 static int do_tcpdrain = 1;
216 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
217 "Enable tcp_drain routine for extra help when low on mbufs");
219 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
220 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
222 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1;
223 #define V_icmp_may_rst VNET(icmp_may_rst)
224 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
225 &VNET_NAME(icmp_may_rst), 0,
226 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
228 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0;
229 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
230 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
231 &VNET_NAME(tcp_isn_reseed_interval), 0,
232 "Seconds between reseeding of ISN secret");
234 static int tcp_soreceive_stream;
235 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
236 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
238 VNET_DEFINE(uma_zone_t, sack_hole_zone);
239 #define V_sack_hole_zone VNET(sack_hole_zone)
242 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
245 #define TS_OFFSET_SECRET_LENGTH 32
246 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]);
247 #define V_ts_offset_secret VNET(ts_offset_secret)
249 static int tcp_default_fb_init(struct tcpcb *tp);
250 static void tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged);
251 static int tcp_default_handoff_ok(struct tcpcb *tp);
252 static struct inpcb *tcp_notify(struct inpcb *, int);
253 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
254 static void tcp_mtudisc(struct inpcb *, int);
255 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
256 void *ip4hdr, const void *ip6hdr);
259 static struct tcp_function_block tcp_def_funcblk = {
260 .tfb_tcp_block_name = "freebsd",
261 .tfb_tcp_output = tcp_output,
262 .tfb_tcp_do_segment = tcp_do_segment,
263 .tfb_tcp_ctloutput = tcp_default_ctloutput,
264 .tfb_tcp_handoff_ok = tcp_default_handoff_ok,
265 .tfb_tcp_fb_init = tcp_default_fb_init,
266 .tfb_tcp_fb_fini = tcp_default_fb_fini,
269 static int tcp_fb_cnt = 0;
270 struct tcp_funchead t_functions;
271 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
273 static struct tcp_function_block *
274 find_tcp_functions_locked(struct tcp_function_set *fs)
276 struct tcp_function *f;
277 struct tcp_function_block *blk=NULL;
279 TAILQ_FOREACH(f, &t_functions, tf_next) {
280 if (strcmp(f->tf_name, fs->function_set_name) == 0) {
288 static struct tcp_function_block *
289 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
291 struct tcp_function_block *rblk=NULL;
292 struct tcp_function *f;
294 TAILQ_FOREACH(f, &t_functions, tf_next) {
295 if (f->tf_fb == blk) {
306 struct tcp_function_block *
307 find_and_ref_tcp_functions(struct tcp_function_set *fs)
309 struct tcp_function_block *blk;
311 rw_rlock(&tcp_function_lock);
312 blk = find_tcp_functions_locked(fs);
314 refcount_acquire(&blk->tfb_refcnt);
315 rw_runlock(&tcp_function_lock);
319 struct tcp_function_block *
320 find_and_ref_tcp_fb(struct tcp_function_block *blk)
322 struct tcp_function_block *rblk;
324 rw_rlock(&tcp_function_lock);
325 rblk = find_tcp_fb_locked(blk, NULL);
327 refcount_acquire(&rblk->tfb_refcnt);
328 rw_runlock(&tcp_function_lock);
332 static struct tcp_function_block *
333 find_and_ref_tcp_default_fb(void)
335 struct tcp_function_block *rblk;
337 rw_rlock(&tcp_function_lock);
338 rblk = tcp_func_set_ptr;
339 refcount_acquire(&rblk->tfb_refcnt);
340 rw_runlock(&tcp_function_lock);
345 tcp_switch_back_to_default(struct tcpcb *tp)
347 struct tcp_function_block *tfb;
349 KASSERT(tp->t_fb != &tcp_def_funcblk,
350 ("%s: called by the built-in default stack", __func__));
353 * Release the old stack. This function will either find a new one
356 if (tp->t_fb->tfb_tcp_fb_fini != NULL)
357 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
358 refcount_release(&tp->t_fb->tfb_refcnt);
361 * Now, we'll find a new function block to use.
362 * Start by trying the current user-selected
363 * default, unless this stack is the user-selected
366 tfb = find_and_ref_tcp_default_fb();
367 if (tfb == tp->t_fb) {
368 refcount_release(&tfb->tfb_refcnt);
371 /* Does the stack accept this connection? */
372 if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL &&
373 (*tfb->tfb_tcp_handoff_ok)(tp)) {
374 refcount_release(&tfb->tfb_refcnt);
377 /* Try to use that stack. */
379 /* Initialize the new stack. If it succeeds, we are done. */
381 if (tp->t_fb->tfb_tcp_fb_init == NULL ||
382 (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0)
386 * Initialization failed. Release the reference count on
389 refcount_release(&tfb->tfb_refcnt);
393 * If that wasn't feasible, use the built-in default
394 * stack which is not allowed to reject anyone.
396 tfb = find_and_ref_tcp_fb(&tcp_def_funcblk);
398 /* there always should be a default */
399 panic("Can't refer to tcp_def_funcblk");
401 if (tfb->tfb_tcp_handoff_ok != NULL) {
402 if ((*tfb->tfb_tcp_handoff_ok) (tp)) {
403 /* The default stack cannot say no */
404 panic("Default stack rejects a new session?");
408 if (tp->t_fb->tfb_tcp_fb_init != NULL &&
409 (*tp->t_fb->tfb_tcp_fb_init)(tp)) {
410 /* The default stack cannot fail */
411 panic("Default stack initialization failed");
416 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
419 struct tcp_function_set fs;
420 struct tcp_function_block *blk;
422 memset(&fs, 0, sizeof(fs));
423 rw_rlock(&tcp_function_lock);
424 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
427 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
428 fs.pcbcnt = blk->tfb_refcnt;
430 rw_runlock(&tcp_function_lock);
431 error = sysctl_handle_string(oidp, fs.function_set_name,
432 sizeof(fs.function_set_name), req);
434 /* Check for error or no change */
435 if (error != 0 || req->newptr == NULL)
438 rw_wlock(&tcp_function_lock);
439 blk = find_tcp_functions_locked(&fs);
441 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
445 tcp_func_set_ptr = blk;
447 rw_wunlock(&tcp_function_lock);
451 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
452 CTLTYPE_STRING | CTLFLAG_RW,
453 NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
454 "Set/get the default TCP functions");
457 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
459 int error, cnt, linesz;
460 struct tcp_function *f;
466 rw_rlock(&tcp_function_lock);
467 TAILQ_FOREACH(f, &t_functions, tf_next) {
470 rw_runlock(&tcp_function_lock);
472 bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
473 buffer = malloc(bufsz, M_TEMP, M_WAITOK);
478 linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
479 "Alias", "PCB count");
484 rw_rlock(&tcp_function_lock);
485 TAILQ_FOREACH(f, &t_functions, tf_next) {
486 alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
487 linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
488 f->tf_fb->tfb_tcp_block_name,
489 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
490 alias ? f->tf_name : "-",
491 f->tf_fb->tfb_refcnt);
492 if (linesz >= bufsz) {
500 rw_runlock(&tcp_function_lock);
502 error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
503 free(buffer, M_TEMP);
507 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
508 CTLTYPE_STRING|CTLFLAG_RD,
509 NULL, 0, sysctl_net_inet_list_available, "A",
510 "list available TCP Function sets");
513 * Exports one (struct tcp_function_info) for each alias/name.
516 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS)
519 struct tcp_function *f;
520 struct tcp_function_info tfi;
523 * We don't allow writes.
525 if (req->newptr != NULL)
529 * Wire the old buffer so we can directly copy the functions to
530 * user space without dropping the lock.
532 if (req->oldptr != NULL) {
533 error = sysctl_wire_old_buffer(req, 0);
539 * Walk the list and copy out matching entries. If INVARIANTS
540 * is compiled in, also walk the list to verify the length of
541 * the list matches what we have recorded.
543 rw_rlock(&tcp_function_lock);
547 if (req->oldptr == NULL) {
552 TAILQ_FOREACH(f, &t_functions, tf_next) {
556 if (req->oldptr != NULL) {
557 bzero(&tfi, sizeof(tfi));
558 tfi.tfi_refcnt = f->tf_fb->tfb_refcnt;
559 tfi.tfi_id = f->tf_fb->tfb_id;
560 (void)strlcpy(tfi.tfi_alias, f->tf_name,
561 sizeof(tfi.tfi_alias));
562 (void)strlcpy(tfi.tfi_name,
563 f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name));
564 error = SYSCTL_OUT(req, &tfi, sizeof(tfi));
566 * Don't stop on error, as that is the
567 * mechanism we use to accumulate length
568 * information if the buffer was too short.
572 KASSERT(cnt == tcp_fb_cnt,
573 ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt));
577 rw_runlock(&tcp_function_lock);
578 if (req->oldptr == NULL)
579 error = SYSCTL_OUT(req, NULL,
580 (cnt + 1) * sizeof(struct tcp_function_info));
585 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info,
586 CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE,
587 NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info",
588 "List TCP function block name-to-ID mappings");
591 * tfb_tcp_handoff_ok() function for the default stack.
592 * Note that we'll basically try to take all comers.
595 tcp_default_handoff_ok(struct tcpcb *tp)
602 * tfb_tcp_fb_init() function for the default stack.
604 * This handles making sure we have appropriate timers set if you are
605 * transitioning a socket that has some amount of setup done.
607 * The init() fuction from the default can *never* return non-zero i.e.
608 * it is required to always succeed since it is the stack of last resort!
611 tcp_default_fb_init(struct tcpcb *tp)
616 INP_WLOCK_ASSERT(tp->t_inpcb);
618 KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT,
619 ("%s: connection %p in unexpected state %d", __func__, tp,
623 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't
624 * know what to do for unexpected states (which includes TIME_WAIT).
626 if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT)
630 * Make sure some kind of transmission timer is set if there is
633 so = tp->t_inpcb->inp_socket;
634 if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) ||
635 tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) ||
636 tcp_timer_active(tp, TT_PERSIST))) {
638 * If the session has established and it looks like it should
639 * be in the persist state, set the persist timer. Otherwise,
640 * set the retransmit timer.
642 if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 &&
643 (int32_t)(tp->snd_nxt - tp->snd_una) <
644 (int32_t)sbavail(&so->so_snd))
647 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
650 /* All non-embryonic sessions get a keepalive timer. */
651 if (!tcp_timer_active(tp, TT_KEEP))
652 tcp_timer_activate(tp, TT_KEEP,
653 TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) :
660 * tfb_tcp_fb_fini() function for the default stack.
662 * This changes state as necessary (or prudent) to prepare for another stack
663 * to assume responsibility for the connection.
666 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged)
669 INP_WLOCK_ASSERT(tp->t_inpcb);
674 * Target size of TCP PCB hash tables. Must be a power of two.
676 * Note that this can be overridden by the kernel environment
677 * variable net.inet.tcp.tcbhashsize
680 #define TCBHASHSIZE 0
685 * Callouts should be moved into struct tcp directly. They are currently
686 * separate because the tcpcb structure is exported to userland for sysctl
687 * parsing purposes, which do not know about callouts.
698 VNET_DEFINE_STATIC(uma_zone_t, tcpcb_zone);
699 #define V_tcpcb_zone VNET(tcpcb_zone)
701 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
702 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
704 static struct mtx isn_mtx;
706 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
707 #define ISN_LOCK() mtx_lock(&isn_mtx)
708 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
711 * TCP initialization.
714 tcp_zone_change(void *tag)
717 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
718 uma_zone_set_max(V_tcpcb_zone, maxsockets);
719 tcp_tw_zone_change();
723 tcp_inpcb_init(void *mem, int size, int flags)
725 struct inpcb *inp = mem;
727 INP_LOCK_INIT(inp, "inp", "tcpinp");
732 * Take a value and get the next power of 2 that doesn't overflow.
733 * Used to size the tcp_inpcb hash buckets.
736 maketcp_hashsize(int size)
742 * get the next power of 2 higher than maxsockets.
744 hashsize = 1 << fls(size);
745 /* catch overflow, and just go one power of 2 smaller */
746 if (hashsize < size) {
747 hashsize = 1 << (fls(size) - 1);
752 static volatile int next_tcp_stack_id = 1;
755 * Register a TCP function block with the name provided in the names
756 * array. (Note that this function does NOT automatically register
757 * blk->tfb_tcp_block_name as a stack name. Therefore, you should
758 * explicitly include blk->tfb_tcp_block_name in the list of names if
759 * you wish to register the stack with that name.)
761 * Either all name registrations will succeed or all will fail. If
762 * a name registration fails, the function will update the num_names
763 * argument to point to the array index of the name that encountered
766 * Returns 0 on success, or an error code on failure.
769 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
770 const char *names[], int *num_names)
772 struct tcp_function *n;
773 struct tcp_function_set fs;
776 KASSERT(names != NULL && *num_names > 0,
777 ("%s: Called with 0-length name list", __func__));
778 KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
779 KASSERT(rw_initialized(&tcp_function_lock),
780 ("%s: called too early", __func__));
782 if ((blk->tfb_tcp_output == NULL) ||
783 (blk->tfb_tcp_do_segment == NULL) ||
784 (blk->tfb_tcp_ctloutput == NULL) ||
785 (strlen(blk->tfb_tcp_block_name) == 0)) {
787 * These functions are required and you
793 if (blk->tfb_tcp_timer_stop_all ||
794 blk->tfb_tcp_timer_activate ||
795 blk->tfb_tcp_timer_active ||
796 blk->tfb_tcp_timer_stop) {
798 * If you define one timer function you
799 * must have them all.
801 if ((blk->tfb_tcp_timer_stop_all == NULL) ||
802 (blk->tfb_tcp_timer_activate == NULL) ||
803 (blk->tfb_tcp_timer_active == NULL) ||
804 (blk->tfb_tcp_timer_stop == NULL)) {
810 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
815 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)strlcpy(fs.function_set_name, names[i],
826 sizeof(fs.function_set_name));
827 rw_wlock(&tcp_function_lock);
828 if (find_tcp_functions_locked(&fs) != NULL) {
829 /* Duplicate name space not allowed */
830 rw_wunlock(&tcp_function_lock);
831 free(n, M_TCPFUNCTIONS);
835 (void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name));
836 TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
838 rw_wunlock(&tcp_function_lock);
844 * Deregister the names we just added. Because registration failed
845 * for names[i], we don't need to deregister that name.
848 rw_wlock(&tcp_function_lock);
850 TAILQ_FOREACH(n, &t_functions, tf_next) {
851 if (!strncmp(n->tf_name, names[i],
852 TCP_FUNCTION_NAME_LEN_MAX)) {
853 TAILQ_REMOVE(&t_functions, n, tf_next);
856 free(n, M_TCPFUNCTIONS);
861 rw_wunlock(&tcp_function_lock);
866 * Register a TCP function block using the name provided in the name
869 * Returns 0 on success, or an error code on failure.
872 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
875 const char *name_list[1];
882 name_list[0] = blk->tfb_tcp_block_name;
883 rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
888 * Register a TCP function block using the name defined in
889 * blk->tfb_tcp_block_name.
891 * Returns 0 on success, or an error code on failure.
894 register_tcp_functions(struct tcp_function_block *blk, int wait)
897 return (register_tcp_functions_as_name(blk, NULL, wait));
901 * Deregister all names associated with a function block. This
902 * functionally removes the function block from use within the system.
904 * When called with a true quiesce argument, mark the function block
905 * as being removed so no more stacks will use it and determine
906 * whether the removal would succeed.
908 * When called with a false quiesce argument, actually attempt the
911 * When called with a force argument, attempt to switch all TCBs to
912 * use the default stack instead of returning EBUSY.
914 * Returns 0 on success (or if the removal would succeed, or an error
918 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce,
921 struct tcp_function *f;
923 if (blk == &tcp_def_funcblk) {
924 /* You can't un-register the default */
927 rw_wlock(&tcp_function_lock);
928 if (blk == tcp_func_set_ptr) {
929 /* You can't free the current default */
930 rw_wunlock(&tcp_function_lock);
933 /* Mark the block so no more stacks can use it. */
934 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
936 * If TCBs are still attached to the stack, attempt to switch them
937 * to the default stack.
939 if (force && blk->tfb_refcnt) {
942 VNET_ITERATOR_DECL(vnet_iter);
944 rw_wunlock(&tcp_function_lock);
947 VNET_FOREACH(vnet_iter) {
948 CURVNET_SET(vnet_iter);
949 INP_INFO_WLOCK(&V_tcbinfo);
950 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
952 if (inp->inp_flags & INP_TIMEWAIT) {
957 if (tp == NULL || tp->t_fb != blk) {
961 tcp_switch_back_to_default(tp);
964 INP_INFO_WUNLOCK(&V_tcbinfo);
969 rw_wlock(&tcp_function_lock);
971 if (blk->tfb_refcnt) {
972 /* TCBs still attached. */
973 rw_wunlock(&tcp_function_lock);
978 rw_wunlock(&tcp_function_lock);
981 /* Remove any function names that map to this function block. */
982 while (find_tcp_fb_locked(blk, &f) != NULL) {
983 TAILQ_REMOVE(&t_functions, f, tf_next);
986 free(f, M_TCPFUNCTIONS);
988 rw_wunlock(&tcp_function_lock);
995 const char *tcbhash_tuneable;
998 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
1001 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
1002 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1003 printf("%s: WARNING: unable to register helper hook\n", __func__);
1004 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
1005 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1006 printf("%s: WARNING: unable to register helper hook\n", __func__);
1008 hashsize = TCBHASHSIZE;
1009 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
1010 if (hashsize == 0) {
1012 * Auto tune the hash size based on maxsockets.
1013 * A perfect hash would have a 1:1 mapping
1014 * (hashsize = maxsockets) however it's been
1015 * suggested that O(2) average is better.
1017 hashsize = maketcp_hashsize(maxsockets / 4);
1019 * Our historical default is 512,
1020 * do not autotune lower than this.
1024 if (bootverbose && IS_DEFAULT_VNET(curvnet))
1025 printf("%s: %s auto tuned to %d\n", __func__,
1026 tcbhash_tuneable, hashsize);
1029 * We require a hashsize to be a power of two.
1030 * Previously if it was not a power of two we would just reset it
1031 * back to 512, which could be a nasty surprise if you did not notice
1032 * the error message.
1033 * Instead what we do is clip it to the closest power of two lower
1034 * than the specified hash value.
1036 if (!powerof2(hashsize)) {
1037 int oldhashsize = hashsize;
1039 hashsize = maketcp_hashsize(hashsize);
1040 /* prevent absurdly low value */
1043 printf("%s: WARNING: TCB hash size not a power of 2, "
1044 "clipped from %d to %d.\n", __func__, oldhashsize,
1047 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
1048 "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE);
1051 * These have to be type stable for the benefit of the timers.
1053 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
1054 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1055 uma_zone_set_max(V_tcpcb_zone, maxsockets);
1056 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
1062 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
1063 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
1064 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1066 tcp_fastopen_init();
1068 /* Skip initialization of globals for non-default instances. */
1069 if (!IS_DEFAULT_VNET(curvnet))
1072 tcp_reass_global_init();
1074 /* XXX virtualize those bellow? */
1075 tcp_delacktime = TCPTV_DELACK;
1076 tcp_keepinit = TCPTV_KEEP_INIT;
1077 tcp_keepidle = TCPTV_KEEP_IDLE;
1078 tcp_keepintvl = TCPTV_KEEPINTVL;
1079 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
1080 tcp_msl = TCPTV_MSL;
1081 tcp_rexmit_initial = TCPTV_RTOBASE;
1082 if (tcp_rexmit_initial < 1)
1083 tcp_rexmit_initial = 1;
1084 tcp_rexmit_min = TCPTV_MIN;
1085 if (tcp_rexmit_min < 1)
1087 tcp_persmin = TCPTV_PERSMIN;
1088 tcp_persmax = TCPTV_PERSMAX;
1089 tcp_rexmit_slop = TCPTV_CPU_VAR;
1090 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
1091 tcp_tcbhashsize = hashsize;
1093 /* Setup the tcp function block list */
1094 TAILQ_INIT(&t_functions);
1095 rw_init(&tcp_function_lock, "tcp_func_lock");
1096 register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
1098 /* Initialize the TCP logging data. */
1101 arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0);
1103 if (tcp_soreceive_stream) {
1105 tcp_usrreqs.pru_soreceive = soreceive_stream;
1108 tcp6_usrreqs.pru_soreceive = soreceive_stream;
1113 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
1115 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
1117 if (max_protohdr < TCP_MINPROTOHDR)
1118 max_protohdr = TCP_MINPROTOHDR;
1119 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
1121 #undef TCP_MINPROTOHDR
1124 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
1125 SHUTDOWN_PRI_DEFAULT);
1126 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
1127 EVENTHANDLER_PRI_ANY);
1135 tcp_destroy(void *unused __unused)
1143 * All our processes are gone, all our sockets should be cleaned
1144 * up, which means, we should be past the tcp_discardcb() calls.
1145 * Sleep to let all tcpcb timers really disappear and cleanup.
1148 INP_LIST_RLOCK(&V_tcbinfo);
1149 n = V_tcbinfo.ipi_count;
1150 INP_LIST_RUNLOCK(&V_tcbinfo);
1153 pause("tcpdes", hz / 10);
1158 in_pcbinfo_destroy(&V_tcbinfo);
1159 /* tcp_discardcb() clears the sack_holes up. */
1160 uma_zdestroy(V_sack_hole_zone);
1161 uma_zdestroy(V_tcpcb_zone);
1164 * Cannot free the zone until all tcpcbs are released as we attach
1165 * the allocations to them.
1167 tcp_fastopen_destroy();
1170 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
1172 printf("%s: WARNING: unable to deregister helper hook "
1173 "type=%d, id=%d: error %d returned\n", __func__,
1174 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
1176 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
1178 printf("%s: WARNING: unable to deregister helper hook "
1179 "type=%d, id=%d: error %d returned\n", __func__,
1180 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
1184 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
1194 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
1195 * tcp_template used to store this data in mbufs, but we now recopy it out
1196 * of the tcpcb each time to conserve mbufs.
1199 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
1201 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
1203 INP_WLOCK_ASSERT(inp);
1206 if ((inp->inp_vflag & INP_IPV6) != 0) {
1207 struct ip6_hdr *ip6;
1209 ip6 = (struct ip6_hdr *)ip_ptr;
1210 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
1211 (inp->inp_flow & IPV6_FLOWINFO_MASK);
1212 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
1213 (IPV6_VERSION & IPV6_VERSION_MASK);
1214 ip6->ip6_nxt = IPPROTO_TCP;
1215 ip6->ip6_plen = htons(sizeof(struct tcphdr));
1216 ip6->ip6_src = inp->in6p_laddr;
1217 ip6->ip6_dst = inp->in6p_faddr;
1220 #if defined(INET6) && defined(INET)
1227 ip = (struct ip *)ip_ptr;
1228 ip->ip_v = IPVERSION;
1230 ip->ip_tos = inp->inp_ip_tos;
1234 ip->ip_ttl = inp->inp_ip_ttl;
1236 ip->ip_p = IPPROTO_TCP;
1237 ip->ip_src = inp->inp_laddr;
1238 ip->ip_dst = inp->inp_faddr;
1241 th->th_sport = inp->inp_lport;
1242 th->th_dport = inp->inp_fport;
1250 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
1254 * Create template to be used to send tcp packets on a connection.
1255 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
1256 * use for this function is in keepalives, which use tcp_respond.
1259 tcpip_maketemplate(struct inpcb *inp)
1263 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
1266 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
1271 * Send a single message to the TCP at address specified by
1272 * the given TCP/IP header. If m == NULL, then we make a copy
1273 * of the tcpiphdr at th and send directly to the addressed host.
1274 * This is used to force keep alive messages out using the TCP
1275 * template for a connection. If flags are given then we send
1276 * a message back to the TCP which originated the segment th,
1277 * and discard the mbuf containing it and any other attached mbufs.
1279 * In any case the ack and sequence number of the transmitted
1280 * segment are as specified by the parameters.
1282 * NOTE: If m != NULL, then th must point to *inside* the mbuf.
1285 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
1286 tcp_seq ack, tcp_seq seq, int flags)
1295 struct ip6_hdr *ip6;
1298 int optlen, tlen, win;
1301 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
1304 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
1311 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
1312 INP_WLOCK_ASSERT(inp);
1319 if (!(flags & TH_RST)) {
1320 win = sbspace(&inp->inp_socket->so_rcv);
1321 if (win > TCP_MAXWIN << tp->rcv_scale)
1322 win = TCP_MAXWIN << tp->rcv_scale;
1324 if ((tp->t_flags & TF_NOOPT) == 0)
1328 m = m_gethdr(M_NOWAIT, MT_DATA);
1331 m->m_data += max_linkhdr;
1334 bcopy((caddr_t)ip6, mtod(m, caddr_t),
1335 sizeof(struct ip6_hdr));
1336 ip6 = mtod(m, struct ip6_hdr *);
1337 nth = (struct tcphdr *)(ip6 + 1);
1341 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1342 ip = mtod(m, struct ip *);
1343 nth = (struct tcphdr *)(ip + 1);
1345 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1347 } else if (!M_WRITABLE(m)) {
1350 /* Can't reuse 'm', allocate a new mbuf. */
1351 n = m_gethdr(M_NOWAIT, MT_DATA);
1357 if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1363 n->m_data += max_linkhdr;
1364 /* m_len is set later */
1365 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1368 bcopy((caddr_t)ip6, mtod(n, caddr_t),
1369 sizeof(struct ip6_hdr));
1370 ip6 = mtod(n, struct ip6_hdr *);
1371 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1372 nth = (struct tcphdr *)(ip6 + 1);
1376 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1377 ip = mtod(n, struct ip *);
1378 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1379 nth = (struct tcphdr *)(ip + 1);
1381 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1382 xchg(nth->th_dport, nth->th_sport, uint16_t);
1389 * XXX MRT We inherit the FIB, which is lucky.
1393 m->m_data = (caddr_t)ipgen;
1394 /* m_len is set later */
1397 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1398 nth = (struct tcphdr *)(ip6 + 1);
1402 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1403 nth = (struct tcphdr *)(ip + 1);
1407 * this is usually a case when an extension header
1408 * exists between the IPv6 header and the
1411 nth->th_sport = th->th_sport;
1412 nth->th_dport = th->th_dport;
1414 xchg(nth->th_dport, nth->th_sport, uint16_t);
1420 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1422 #if defined(INET) && defined(INET6)
1426 tlen = sizeof (struct tcpiphdr);
1430 KASSERT(M_TRAILINGSPACE(m) >= tlen,
1431 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1432 m, tlen, (long)M_TRAILINGSPACE(m)));
1437 /* Make sure we have room. */
1438 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1439 m->m_next = m_get(M_NOWAIT, MT_DATA);
1441 optp = mtod(m->m_next, u_char *);
1446 optp = (u_char *) (nth + 1);
1452 if (tp->t_flags & TF_RCVD_TSTMP) {
1453 to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1454 to.to_tsecr = tp->ts_recent;
1455 to.to_flags |= TOF_TS;
1457 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1458 /* TCP-MD5 (RFC2385). */
1459 if (tp->t_flags & TF_SIGNATURE)
1460 to.to_flags |= TOF_SIGNATURE;
1462 /* Add the options. */
1463 tlen += optlen = tcp_addoptions(&to, optp);
1465 /* Update m_len in the correct mbuf. */
1466 optm->m_len += optlen;
1472 ip6->ip6_vfc = IPV6_VERSION;
1473 ip6->ip6_nxt = IPPROTO_TCP;
1474 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1477 #if defined(INET) && defined(INET6)
1482 ip->ip_len = htons(tlen);
1483 ip->ip_ttl = V_ip_defttl;
1484 if (V_path_mtu_discovery)
1485 ip->ip_off |= htons(IP_DF);
1488 m->m_pkthdr.len = tlen;
1489 m->m_pkthdr.rcvif = NULL;
1493 * Packet is associated with a socket, so allow the
1494 * label of the response to reflect the socket label.
1496 INP_WLOCK_ASSERT(inp);
1497 mac_inpcb_create_mbuf(inp, m);
1500 * Packet is not associated with a socket, so possibly
1501 * update the label in place.
1503 mac_netinet_tcp_reply(m);
1506 nth->th_seq = htonl(seq);
1507 nth->th_ack = htonl(ack);
1509 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1510 nth->th_flags = flags;
1512 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1514 nth->th_win = htons((u_short)win);
1517 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1518 if (to.to_flags & TOF_SIGNATURE) {
1519 if (!TCPMD5_ENABLED() ||
1520 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
1527 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1530 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1531 nth->th_sum = in6_cksum_pseudo(ip6,
1532 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
1533 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
1537 #if defined(INET6) && defined(INET)
1542 m->m_pkthdr.csum_flags = CSUM_TCP;
1543 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1544 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1548 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1549 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1551 TCP_PROBE3(debug__output, tp, th, m);
1553 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
1557 TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
1558 (void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
1561 #if defined(INET) && defined(INET6)
1566 TCP_PROBE5(send, NULL, tp, ip, tp, nth);
1567 (void)ip_output(m, NULL, NULL, 0, NULL, inp);
1573 * Create a new TCP control block, making an
1574 * empty reassembly queue and hooking it to the argument
1575 * protocol control block. The `inp' parameter must have
1576 * come from the zone allocator set up in tcp_init().
1579 tcp_newtcpcb(struct inpcb *inp)
1581 struct tcpcb_mem *tm;
1584 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1587 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1592 /* Initialise cc_var struct for this tcpcb. */
1594 tp->ccv->type = IPPROTO_TCP;
1595 tp->ccv->ccvc.tcp = tp;
1596 rw_rlock(&tcp_function_lock);
1597 tp->t_fb = tcp_func_set_ptr;
1598 refcount_acquire(&tp->t_fb->tfb_refcnt);
1599 rw_runlock(&tcp_function_lock);
1601 * Use the current system default CC algorithm.
1604 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1605 CC_ALGO(tp) = CC_DEFAULT();
1608 if (CC_ALGO(tp)->cb_init != NULL)
1609 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1610 if (tp->t_fb->tfb_tcp_fb_fini)
1611 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1612 refcount_release(&tp->t_fb->tfb_refcnt);
1613 uma_zfree(V_tcpcb_zone, tm);
1619 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1620 if (tp->t_fb->tfb_tcp_fb_fini)
1621 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1622 refcount_release(&tp->t_fb->tfb_refcnt);
1623 uma_zfree(V_tcpcb_zone, tm);
1629 tp->t_vnet = inp->inp_vnet;
1631 tp->t_timers = &tm->tt;
1632 TAILQ_INIT(&tp->t_segq);
1635 isipv6 ? V_tcp_v6mssdflt :
1639 /* Set up our timeouts. */
1640 callout_init(&tp->t_timers->tt_rexmt, 1);
1641 callout_init(&tp->t_timers->tt_persist, 1);
1642 callout_init(&tp->t_timers->tt_keep, 1);
1643 callout_init(&tp->t_timers->tt_2msl, 1);
1644 callout_init(&tp->t_timers->tt_delack, 1);
1646 if (V_tcp_do_rfc1323)
1647 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1649 tp->t_flags |= TF_SACK_PERMIT;
1650 TAILQ_INIT(&tp->snd_holes);
1652 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1655 in_pcbref(inp); /* Reference for tcpcb */
1659 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1660 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
1661 * reasonable initial retransmit time.
1663 tp->t_srtt = TCPTV_SRTTBASE;
1664 tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1665 tp->t_rttmin = tcp_rexmit_min;
1666 tp->t_rxtcur = tcp_rexmit_initial;
1667 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1668 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1669 tp->t_rcvtime = ticks;
1671 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1672 * because the socket may be bound to an IPv6 wildcard address,
1673 * which may match an IPv4-mapped IPv6 address.
1675 inp->inp_ip_ttl = V_ip_defttl;
1679 * Init the TCP PCAP queues.
1681 tcp_pcap_tcpcb_init(tp);
1684 /* Initialize the per-TCPCB log data. */
1685 tcp_log_tcpcbinit(tp);
1687 if (tp->t_fb->tfb_tcp_fb_init) {
1688 (*tp->t_fb->tfb_tcp_fb_init)(tp);
1690 return (tp); /* XXX */
1694 * Switch the congestion control algorithm back to NewReno for any active
1695 * control blocks using an algorithm which is about to go away.
1696 * This ensures the CC framework can allow the unload to proceed without leaving
1697 * any dangling pointers which would trigger a panic.
1698 * Returning non-zero would inform the CC framework that something went wrong
1699 * and it would be unsafe to allow the unload to proceed. However, there is no
1700 * way for this to occur with this implementation so we always return zero.
1703 tcp_ccalgounload(struct cc_algo *unload_algo)
1705 struct cc_algo *tmpalgo;
1708 VNET_ITERATOR_DECL(vnet_iter);
1711 * Check all active control blocks across all network stacks and change
1712 * any that are using "unload_algo" back to NewReno. If "unload_algo"
1713 * requires cleanup code to be run, call it.
1716 VNET_FOREACH(vnet_iter) {
1717 CURVNET_SET(vnet_iter);
1718 INP_INFO_WLOCK(&V_tcbinfo);
1720 * New connections already part way through being initialised
1721 * with the CC algo we're removing will not race with this code
1722 * because the INP_INFO_WLOCK is held during initialisation. We
1723 * therefore don't enter the loop below until the connection
1724 * list has stabilised.
1726 CK_LIST_FOREACH(inp, &V_tcb, inp_list) {
1728 /* Important to skip tcptw structs. */
1729 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1730 (tp = intotcpcb(inp)) != NULL) {
1732 * By holding INP_WLOCK here, we are assured
1733 * that the connection is not currently
1734 * executing inside the CC module's functions
1735 * i.e. it is safe to make the switch back to
1738 if (CC_ALGO(tp) == unload_algo) {
1739 tmpalgo = CC_ALGO(tp);
1740 if (tmpalgo->cb_destroy != NULL)
1741 tmpalgo->cb_destroy(tp->ccv);
1744 * NewReno may allocate memory on
1745 * demand for certain stateful
1746 * configuration as needed, but is
1747 * coded to never fail on memory
1748 * allocation failure so it is a safe
1751 CC_ALGO(tp) = &newreno_cc_algo;
1756 INP_INFO_WUNLOCK(&V_tcbinfo);
1759 VNET_LIST_RUNLOCK();
1765 * Drop a TCP connection, reporting
1766 * the specified error. If connection is synchronized,
1767 * then send a RST to peer.
1770 tcp_drop(struct tcpcb *tp, int errno)
1772 struct socket *so = tp->t_inpcb->inp_socket;
1774 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1775 INP_WLOCK_ASSERT(tp->t_inpcb);
1777 if (TCPS_HAVERCVDSYN(tp->t_state)) {
1778 tcp_state_change(tp, TCPS_CLOSED);
1779 (void) tp->t_fb->tfb_tcp_output(tp);
1780 TCPSTAT_INC(tcps_drops);
1782 TCPSTAT_INC(tcps_conndrops);
1783 if (errno == ETIMEDOUT && tp->t_softerror)
1784 errno = tp->t_softerror;
1785 so->so_error = errno;
1786 return (tcp_close(tp));
1790 tcp_discardcb(struct tcpcb *tp)
1792 struct inpcb *inp = tp->t_inpcb;
1793 struct socket *so = inp->inp_socket;
1795 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1797 int released __unused;
1799 INP_WLOCK_ASSERT(inp);
1802 * Make sure that all of our timers are stopped before we delete the
1805 * If stopping a timer fails, we schedule a discard function in same
1806 * callout, and the last discard function called will take care of
1807 * deleting the tcpcb.
1809 tp->t_timers->tt_draincnt = 0;
1810 tcp_timer_stop(tp, TT_REXMT);
1811 tcp_timer_stop(tp, TT_PERSIST);
1812 tcp_timer_stop(tp, TT_KEEP);
1813 tcp_timer_stop(tp, TT_2MSL);
1814 tcp_timer_stop(tp, TT_DELACK);
1815 if (tp->t_fb->tfb_tcp_timer_stop_all) {
1817 * Call the stop-all function of the methods,
1818 * this function should call the tcp_timer_stop()
1819 * method with each of the function specific timeouts.
1820 * That stop will be called via the tfb_tcp_timer_stop()
1821 * which should use the async drain function of the
1822 * callout system (see tcp_var.h).
1824 tp->t_fb->tfb_tcp_timer_stop_all(tp);
1828 * If we got enough samples through the srtt filter,
1829 * save the rtt and rttvar in the routing entry.
1830 * 'Enough' is arbitrarily defined as 4 rtt samples.
1831 * 4 samples is enough for the srtt filter to converge
1832 * to within enough % of the correct value; fewer samples
1833 * and we could save a bogus rtt. The danger is not high
1834 * as tcp quickly recovers from everything.
1835 * XXX: Works very well but needs some more statistics!
1837 if (tp->t_rttupdated >= 4) {
1838 struct hc_metrics_lite metrics;
1841 bzero(&metrics, sizeof(metrics));
1843 * Update the ssthresh always when the conditions below
1844 * are satisfied. This gives us better new start value
1845 * for the congestion avoidance for new connections.
1846 * ssthresh is only set if packet loss occurred on a session.
1848 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1849 * being torn down. Ideally this code would not use 'so'.
1851 ssthresh = tp->snd_ssthresh;
1852 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1854 * convert the limit from user data bytes to
1855 * packets then to packet data bytes.
1857 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1860 ssthresh *= (tp->t_maxseg +
1862 (isipv6 ? sizeof (struct ip6_hdr) +
1863 sizeof (struct tcphdr) :
1865 sizeof (struct tcpiphdr)
1872 metrics.rmx_ssthresh = ssthresh;
1874 metrics.rmx_rtt = tp->t_srtt;
1875 metrics.rmx_rttvar = tp->t_rttvar;
1876 metrics.rmx_cwnd = tp->snd_cwnd;
1877 metrics.rmx_sendpipe = 0;
1878 metrics.rmx_recvpipe = 0;
1880 tcp_hc_update(&inp->inp_inc, &metrics);
1883 /* free the reassembly queue, if any */
1884 tcp_reass_flush(tp);
1887 /* Disconnect offload device, if any. */
1888 if (tp->t_flags & TF_TOE)
1889 tcp_offload_detach(tp);
1892 tcp_free_sackholes(tp);
1895 /* Free the TCP PCAP queues. */
1896 tcp_pcap_drain(&(tp->t_inpkts));
1897 tcp_pcap_drain(&(tp->t_outpkts));
1900 /* Allow the CC algorithm to clean up after itself. */
1901 if (CC_ALGO(tp)->cb_destroy != NULL)
1902 CC_ALGO(tp)->cb_destroy(tp->ccv);
1906 khelp_destroy_osd(tp->osd);
1910 inp->inp_ppcb = NULL;
1911 if (tp->t_timers->tt_draincnt == 0) {
1912 /* We own the last reference on tcpcb, let's free it. */
1914 tcp_log_tcpcbfini(tp);
1916 TCPSTATES_DEC(tp->t_state);
1917 if (tp->t_fb->tfb_tcp_fb_fini)
1918 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1919 refcount_release(&tp->t_fb->tfb_refcnt);
1921 uma_zfree(V_tcpcb_zone, tp);
1922 released = in_pcbrele_wlocked(inp);
1923 KASSERT(!released, ("%s: inp %p should not have been released "
1924 "here", __func__, inp));
1929 tcp_timer_discard(void *ptp)
1933 struct epoch_tracker et;
1935 tp = (struct tcpcb *)ptp;
1936 CURVNET_SET(tp->t_vnet);
1937 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
1939 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1942 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1943 ("%s: tcpcb has to be stopped here", __func__));
1944 tp->t_timers->tt_draincnt--;
1945 if (tp->t_timers->tt_draincnt == 0) {
1946 /* We own the last reference on this tcpcb, let's free it. */
1948 tcp_log_tcpcbfini(tp);
1950 TCPSTATES_DEC(tp->t_state);
1951 if (tp->t_fb->tfb_tcp_fb_fini)
1952 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1953 refcount_release(&tp->t_fb->tfb_refcnt);
1955 uma_zfree(V_tcpcb_zone, tp);
1956 if (in_pcbrele_wlocked(inp)) {
1957 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1963 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1968 * Attempt to close a TCP control block, marking it as dropped, and freeing
1969 * the socket if we hold the only reference.
1972 tcp_close(struct tcpcb *tp)
1974 struct inpcb *inp = tp->t_inpcb;
1977 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1978 INP_WLOCK_ASSERT(inp);
1981 if (tp->t_state == TCPS_LISTEN)
1982 tcp_offload_listen_stop(tp);
1985 * This releases the TFO pending counter resource for TFO listen
1986 * sockets as well as passively-created TFO sockets that transition
1987 * from SYN_RECEIVED to CLOSED.
1989 if (tp->t_tfo_pending) {
1990 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
1991 tp->t_tfo_pending = NULL;
1994 TCPSTAT_INC(tcps_closed);
1995 if (tp->t_state != TCPS_CLOSED)
1996 tcp_state_change(tp, TCPS_CLOSED);
1997 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1998 so = inp->inp_socket;
1999 soisdisconnected(so);
2000 if (inp->inp_flags & INP_SOCKREF) {
2001 KASSERT(so->so_state & SS_PROTOREF,
2002 ("tcp_close: !SS_PROTOREF"));
2003 inp->inp_flags &= ~INP_SOCKREF;
2006 so->so_state &= ~SS_PROTOREF;
2016 VNET_ITERATOR_DECL(vnet_iter);
2021 VNET_LIST_RLOCK_NOSLEEP();
2022 VNET_FOREACH(vnet_iter) {
2023 CURVNET_SET(vnet_iter);
2028 * Walk the tcpbs, if existing, and flush the reassembly queue,
2029 * if there is one...
2030 * XXX: The "Net/3" implementation doesn't imply that the TCP
2031 * reassembly queue should be flushed, but in a situation
2032 * where we're really low on mbufs, this is potentially
2035 INP_INFO_WLOCK(&V_tcbinfo);
2036 CK_LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
2038 if (inpb->inp_flags & INP_TIMEWAIT) {
2042 if ((tcpb = intotcpcb(inpb)) != NULL) {
2043 tcp_reass_flush(tcpb);
2044 tcp_clean_sackreport(tcpb);
2046 tcp_log_drain(tcpb);
2049 if (tcp_pcap_aggressive_free) {
2050 /* Free the TCP PCAP queues. */
2051 tcp_pcap_drain(&(tcpb->t_inpkts));
2052 tcp_pcap_drain(&(tcpb->t_outpkts));
2058 INP_INFO_WUNLOCK(&V_tcbinfo);
2061 VNET_LIST_RUNLOCK_NOSLEEP();
2065 * Notify a tcp user of an asynchronous error;
2066 * store error as soft error, but wake up user
2067 * (for now, won't do anything until can select for soft error).
2069 * Do not wake up user since there currently is no mechanism for
2070 * reporting soft errors (yet - a kqueue filter may be added).
2072 static struct inpcb *
2073 tcp_notify(struct inpcb *inp, int error)
2077 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2078 INP_WLOCK_ASSERT(inp);
2080 if ((inp->inp_flags & INP_TIMEWAIT) ||
2081 (inp->inp_flags & INP_DROPPED))
2084 tp = intotcpcb(inp);
2085 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
2088 * Ignore some errors if we are hooked up.
2089 * If connection hasn't completed, has retransmitted several times,
2090 * and receives a second error, give up now. This is better
2091 * than waiting a long time to establish a connection that
2092 * can never complete.
2094 if (tp->t_state == TCPS_ESTABLISHED &&
2095 (error == EHOSTUNREACH || error == ENETUNREACH ||
2096 error == EHOSTDOWN)) {
2097 if (inp->inp_route.ro_rt) {
2098 RTFREE(inp->inp_route.ro_rt);
2099 inp->inp_route.ro_rt = (struct rtentry *)NULL;
2102 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
2104 tp = tcp_drop(tp, error);
2110 tp->t_softerror = error;
2114 wakeup( &so->so_timeo);
2121 tcp_pcblist(SYSCTL_HANDLER_ARGS)
2123 int error, i, m, n, pcb_count;
2124 struct inpcb *inp, **inp_list;
2127 struct epoch_tracker et;
2130 * The process of preparing the TCB list is too time-consuming and
2131 * resource-intensive to repeat twice on every request.
2133 if (req->oldptr == NULL) {
2134 n = V_tcbinfo.ipi_count +
2135 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2136 n += imax(n / 8, 10);
2137 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
2141 if (req->newptr != NULL)
2145 * OK, now we're committed to doing something.
2147 INP_LIST_RLOCK(&V_tcbinfo);
2148 gencnt = V_tcbinfo.ipi_gencnt;
2149 n = V_tcbinfo.ipi_count;
2150 INP_LIST_RUNLOCK(&V_tcbinfo);
2152 m = counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2154 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
2155 + (n + m) * sizeof(struct xtcpcb));
2159 bzero(&xig, sizeof(xig));
2160 xig.xig_len = sizeof xig;
2161 xig.xig_count = n + m;
2162 xig.xig_gen = gencnt;
2163 xig.xig_sogen = so_gencnt;
2164 error = SYSCTL_OUT(req, &xig, sizeof xig);
2168 error = syncache_pcblist(req, m, &pcb_count);
2172 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
2174 INP_INFO_WLOCK(&V_tcbinfo);
2175 for (inp = CK_LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
2176 inp != NULL && i < n; inp = CK_LIST_NEXT(inp, inp_list)) {
2178 if (inp->inp_gencnt <= gencnt) {
2180 * XXX: This use of cr_cansee(), introduced with
2181 * TCP state changes, is not quite right, but for
2182 * now, better than nothing.
2184 if (inp->inp_flags & INP_TIMEWAIT) {
2185 if (intotw(inp) != NULL)
2186 error = cr_cansee(req->td->td_ucred,
2187 intotw(inp)->tw_cred);
2189 error = EINVAL; /* Skip this inp. */
2191 error = cr_canseeinpcb(req->td->td_ucred, inp);
2194 inp_list[i++] = inp;
2199 INP_INFO_WUNLOCK(&V_tcbinfo);
2203 for (i = 0; i < n; i++) {
2206 if (inp->inp_gencnt <= gencnt) {
2209 tcp_inptoxtp(inp, &xt);
2211 error = SYSCTL_OUT(req, &xt, sizeof xt);
2215 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
2216 for (i = 0; i < n; i++) {
2219 if (!in_pcbrele_rlocked(inp))
2222 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2226 * Give the user an updated idea of our state.
2227 * If the generation differs from what we told
2228 * her before, she knows that something happened
2229 * while we were processing this request, and it
2230 * might be necessary to retry.
2232 INP_LIST_RLOCK(&V_tcbinfo);
2233 xig.xig_gen = V_tcbinfo.ipi_gencnt;
2234 xig.xig_sogen = so_gencnt;
2235 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
2236 INP_LIST_RUNLOCK(&V_tcbinfo);
2237 error = SYSCTL_OUT(req, &xig, sizeof xig);
2239 free(inp_list, M_TEMP);
2243 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
2244 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
2245 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
2249 tcp_getcred(SYSCTL_HANDLER_ARGS)
2252 struct sockaddr_in addrs[2];
2256 error = priv_check(req->td, PRIV_NETINET_GETCRED);
2259 error = SYSCTL_IN(req, addrs, sizeof(addrs));
2262 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
2263 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
2265 if (inp->inp_socket == NULL)
2268 error = cr_canseeinpcb(req->td->td_ucred, inp);
2270 cru2x(inp->inp_cred, &xuc);
2275 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2279 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
2280 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
2281 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
2286 tcp6_getcred(SYSCTL_HANDLER_ARGS)
2289 struct sockaddr_in6 addrs[2];
2296 error = priv_check(req->td, PRIV_NETINET_GETCRED);
2299 error = SYSCTL_IN(req, addrs, sizeof(addrs));
2302 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
2303 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
2306 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
2308 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
2317 inp = in_pcblookup(&V_tcbinfo,
2318 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
2320 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
2321 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
2324 inp = in6_pcblookup(&V_tcbinfo,
2325 &addrs[1].sin6_addr, addrs[1].sin6_port,
2326 &addrs[0].sin6_addr, addrs[0].sin6_port,
2327 INPLOOKUP_RLOCKPCB, NULL);
2329 if (inp->inp_socket == NULL)
2332 error = cr_canseeinpcb(req->td->td_ucred, inp);
2334 cru2x(inp->inp_cred, &xuc);
2339 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2343 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
2344 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
2345 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
2351 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
2353 struct ip *ip = vip;
2355 struct in_addr faddr;
2358 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2360 struct in_conninfo inc;
2361 struct epoch_tracker et;
2362 tcp_seq icmp_tcp_seq;
2365 faddr = ((struct sockaddr_in *)sa)->sin_addr;
2366 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
2369 if (cmd == PRC_MSGSIZE)
2370 notify = tcp_mtudisc_notify;
2371 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2372 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2373 cmd == PRC_TIMXCEED_INTRANS) && ip)
2374 notify = tcp_drop_syn_sent;
2377 * Hostdead is ugly because it goes linearly through all PCBs.
2378 * XXX: We never get this from ICMP, otherwise it makes an
2379 * excellent DoS attack on machines with many connections.
2381 else if (cmd == PRC_HOSTDEAD)
2383 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
2387 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
2391 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
2392 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2393 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
2394 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
2395 th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2396 if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2397 /* signal EHOSTDOWN, as it flushes the cached route */
2398 inp = (*notify)(inp, EHOSTDOWN);
2401 icmp_tcp_seq = th->th_seq;
2403 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2404 !(inp->inp_flags & INP_DROPPED) &&
2405 !(inp->inp_socket == NULL)) {
2406 tp = intotcpcb(inp);
2407 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2408 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2409 if (cmd == PRC_MSGSIZE) {
2412 * If we got a needfrag set the MTU
2413 * in the route to the suggested new
2414 * value (if given) and then notify.
2416 mtu = ntohs(icp->icmp_nextmtu);
2418 * If no alternative MTU was
2419 * proposed, try the next smaller
2424 ntohs(ip->ip_len), 1);
2425 if (mtu < V_tcp_minmss +
2426 sizeof(struct tcpiphdr))
2427 mtu = V_tcp_minmss +
2428 sizeof(struct tcpiphdr);
2430 * Only process the offered MTU if it
2431 * is smaller than the current one.
2433 if (mtu < tp->t_maxseg +
2434 sizeof(struct tcpiphdr)) {
2435 bzero(&inc, sizeof(inc));
2436 inc.inc_faddr = faddr;
2438 inp->inp_inc.inc_fibnum;
2439 tcp_hc_updatemtu(&inc, mtu);
2440 tcp_mtudisc(inp, mtu);
2443 inp = (*notify)(inp,
2444 inetctlerrmap[cmd]);
2448 bzero(&inc, sizeof(inc));
2449 inc.inc_fport = th->th_dport;
2450 inc.inc_lport = th->th_sport;
2451 inc.inc_faddr = faddr;
2452 inc.inc_laddr = ip->ip_src;
2453 syncache_unreach(&inc, icmp_tcp_seq);
2458 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2464 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
2466 struct in6_addr *dst;
2467 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2468 struct ip6_hdr *ip6;
2472 struct icmp6_hdr *icmp6;
2473 struct ip6ctlparam *ip6cp = NULL;
2474 const struct sockaddr_in6 *sa6_src = NULL;
2475 struct in_conninfo inc;
2476 struct epoch_tracker et;
2481 tcp_seq icmp_tcp_seq;
2485 if (sa->sa_family != AF_INET6 ||
2486 sa->sa_len != sizeof(struct sockaddr_in6))
2489 /* if the parameter is from icmp6, decode it. */
2491 ip6cp = (struct ip6ctlparam *)d;
2492 icmp6 = ip6cp->ip6c_icmp6;
2494 ip6 = ip6cp->ip6c_ip6;
2495 off = ip6cp->ip6c_off;
2496 sa6_src = ip6cp->ip6c_src;
2497 dst = ip6cp->ip6c_finaldst;
2501 off = 0; /* fool gcc */
2506 if (cmd == PRC_MSGSIZE)
2507 notify = tcp_mtudisc_notify;
2508 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2509 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2510 cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
2511 notify = tcp_drop_syn_sent;
2514 * Hostdead is ugly because it goes linearly through all PCBs.
2515 * XXX: We never get this from ICMP, otherwise it makes an
2516 * excellent DoS attack on machines with many connections.
2518 else if (cmd == PRC_HOSTDEAD)
2520 else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
2524 in6_pcbnotify(&V_tcbinfo, sa, 0,
2525 (const struct sockaddr *)sa6_src,
2526 0, cmd, NULL, notify);
2530 /* Check if we can safely get the ports from the tcp hdr */
2533 (int32_t) (off + sizeof(struct tcp_ports)))) {
2536 bzero(&t_ports, sizeof(struct tcp_ports));
2537 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
2538 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
2539 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
2540 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
2541 if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2542 /* signal EHOSTDOWN, as it flushes the cached route */
2543 inp = (*notify)(inp, EHOSTDOWN);
2546 off += sizeof(struct tcp_ports);
2547 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
2550 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
2552 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2553 !(inp->inp_flags & INP_DROPPED) &&
2554 !(inp->inp_socket == NULL)) {
2555 tp = intotcpcb(inp);
2556 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2557 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2558 if (cmd == PRC_MSGSIZE) {
2561 * If we got a needfrag set the MTU
2562 * in the route to the suggested new
2563 * value (if given) and then notify.
2565 mtu = ntohl(icmp6->icmp6_mtu);
2567 * If no alternative MTU was
2568 * proposed, or the proposed
2569 * MTU was too small, set to
2572 if (mtu < IPV6_MMTU)
2573 mtu = IPV6_MMTU - 8;
2574 bzero(&inc, sizeof(inc));
2575 inc.inc_fibnum = M_GETFIB(m);
2576 inc.inc_flags |= INC_ISIPV6;
2577 inc.inc6_faddr = *dst;
2578 if (in6_setscope(&inc.inc6_faddr,
2579 m->m_pkthdr.rcvif, NULL))
2582 * Only process the offered MTU if it
2583 * is smaller than the current one.
2585 if (mtu < tp->t_maxseg +
2586 sizeof (struct tcphdr) +
2587 sizeof (struct ip6_hdr)) {
2588 tcp_hc_updatemtu(&inc, mtu);
2589 tcp_mtudisc(inp, mtu);
2590 ICMP6STAT_INC(icp6s_pmtuchg);
2593 inp = (*notify)(inp,
2594 inet6ctlerrmap[cmd]);
2598 bzero(&inc, sizeof(inc));
2599 inc.inc_fibnum = M_GETFIB(m);
2600 inc.inc_flags |= INC_ISIPV6;
2601 inc.inc_fport = t_ports.th_dport;
2602 inc.inc_lport = t_ports.th_sport;
2603 inc.inc6_faddr = *dst;
2604 inc.inc6_laddr = ip6->ip6_src;
2605 syncache_unreach(&inc, icmp_tcp_seq);
2610 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2615 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len)
2621 MD5Update(&ctx, &inc->inc_fport, sizeof(uint16_t));
2622 MD5Update(&ctx, &inc->inc_lport, sizeof(uint16_t));
2623 switch (inc->inc_flags & INC_ISIPV6) {
2626 MD5Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr));
2627 MD5Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr));
2632 MD5Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr));
2633 MD5Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr));
2637 MD5Update(&ctx, key, len);
2638 MD5Final((unsigned char *)hash, &ctx);
2644 tcp_new_ts_offset(struct in_conninfo *inc)
2646 struct in_conninfo inc_store, *local_inc;
2648 if (!V_tcp_ts_offset_per_conn) {
2649 memcpy(&inc_store, inc, sizeof(struct in_conninfo));
2650 inc_store.inc_lport = 0;
2651 inc_store.inc_fport = 0;
2652 local_inc = &inc_store;
2656 return (tcp_keyed_hash(local_inc, V_ts_offset_secret,
2657 sizeof(V_ts_offset_secret)));
2661 * Following is where TCP initial sequence number generation occurs.
2663 * There are two places where we must use initial sequence numbers:
2664 * 1. In SYN-ACK packets.
2665 * 2. In SYN packets.
2667 * All ISNs for SYN-ACK packets are generated by the syncache. See
2668 * tcp_syncache.c for details.
2670 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2671 * depends on this property. In addition, these ISNs should be
2672 * unguessable so as to prevent connection hijacking. To satisfy
2673 * the requirements of this situation, the algorithm outlined in
2674 * RFC 1948 is used, with only small modifications.
2676 * Implementation details:
2678 * Time is based off the system timer, and is corrected so that it
2679 * increases by one megabyte per second. This allows for proper
2680 * recycling on high speed LANs while still leaving over an hour
2683 * As reading the *exact* system time is too expensive to be done
2684 * whenever setting up a TCP connection, we increment the time
2685 * offset in two ways. First, a small random positive increment
2686 * is added to isn_offset for each connection that is set up.
2687 * Second, the function tcp_isn_tick fires once per clock tick
2688 * and increments isn_offset as necessary so that sequence numbers
2689 * are incremented at approximately ISN_BYTES_PER_SECOND. The
2690 * random positive increments serve only to ensure that the same
2691 * exact sequence number is never sent out twice (as could otherwise
2692 * happen when a port is recycled in less than the system tick
2695 * net.inet.tcp.isn_reseed_interval controls the number of seconds
2696 * between seeding of isn_secret. This is normally set to zero,
2697 * as reseeding should not be necessary.
2699 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2700 * isn_offset_old, and isn_ctx is performed using the ISN lock. In
2701 * general, this means holding an exclusive (write) lock.
2704 #define ISN_BYTES_PER_SECOND 1048576
2705 #define ISN_STATIC_INCREMENT 4096
2706 #define ISN_RANDOM_INCREMENT (4096 - 1)
2707 #define ISN_SECRET_LENGTH 32
2709 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]);
2710 VNET_DEFINE_STATIC(int, isn_last);
2711 VNET_DEFINE_STATIC(int, isn_last_reseed);
2712 VNET_DEFINE_STATIC(u_int32_t, isn_offset);
2713 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old);
2715 #define V_isn_secret VNET(isn_secret)
2716 #define V_isn_last VNET(isn_last)
2717 #define V_isn_last_reseed VNET(isn_last_reseed)
2718 #define V_isn_offset VNET(isn_offset)
2719 #define V_isn_offset_old VNET(isn_offset_old)
2722 tcp_new_isn(struct in_conninfo *inc)
2725 u_int32_t projected_offset;
2728 /* Seed if this is the first use, reseed if requested. */
2729 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2730 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2732 arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0);
2733 V_isn_last_reseed = ticks;
2736 /* Compute the md5 hash and return the ISN. */
2737 new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret,
2738 sizeof(V_isn_secret));
2739 V_isn_offset += ISN_STATIC_INCREMENT +
2740 (arc4random() & ISN_RANDOM_INCREMENT);
2741 if (ticks != V_isn_last) {
2742 projected_offset = V_isn_offset_old +
2743 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2744 if (SEQ_GT(projected_offset, V_isn_offset))
2745 V_isn_offset = projected_offset;
2746 V_isn_offset_old = V_isn_offset;
2749 new_isn += V_isn_offset;
2755 * When a specific ICMP unreachable message is received and the
2756 * connection state is SYN-SENT, drop the connection. This behavior
2757 * is controlled by the icmp_may_rst sysctl.
2760 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2764 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2765 INP_WLOCK_ASSERT(inp);
2767 if ((inp->inp_flags & INP_TIMEWAIT) ||
2768 (inp->inp_flags & INP_DROPPED))
2771 tp = intotcpcb(inp);
2772 if (tp->t_state != TCPS_SYN_SENT)
2775 if (IS_FASTOPEN(tp->t_flags))
2776 tcp_fastopen_disable_path(tp);
2778 tp = tcp_drop(tp, errno);
2786 * When `need fragmentation' ICMP is received, update our idea of the MSS
2787 * based on the new value. Also nudge TCP to send something, since we
2788 * know the packet we just sent was dropped.
2789 * This duplicates some code in the tcp_mss() function in tcp_input.c.
2791 static struct inpcb *
2792 tcp_mtudisc_notify(struct inpcb *inp, int error)
2795 tcp_mtudisc(inp, -1);
2800 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2805 INP_WLOCK_ASSERT(inp);
2806 if ((inp->inp_flags & INP_TIMEWAIT) ||
2807 (inp->inp_flags & INP_DROPPED))
2810 tp = intotcpcb(inp);
2811 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2813 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2815 so = inp->inp_socket;
2816 SOCKBUF_LOCK(&so->so_snd);
2817 /* If the mss is larger than the socket buffer, decrease the mss. */
2818 if (so->so_snd.sb_hiwat < tp->t_maxseg)
2819 tp->t_maxseg = so->so_snd.sb_hiwat;
2820 SOCKBUF_UNLOCK(&so->so_snd);
2822 TCPSTAT_INC(tcps_mturesent);
2824 tp->snd_nxt = tp->snd_una;
2825 tcp_free_sackholes(tp);
2826 tp->snd_recover = tp->snd_max;
2827 if (tp->t_flags & TF_SACK_PERMIT)
2828 EXIT_FASTRECOVERY(tp->t_flags);
2829 tp->t_fb->tfb_tcp_output(tp);
2834 * Look-up the routing entry to the peer of this inpcb. If no route
2835 * is found and it cannot be allocated, then return 0. This routine
2836 * is called by TCP routines that access the rmx structure and by
2837 * tcp_mss_update to get the peer/interface MTU.
2840 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2842 struct nhop4_extended nh4;
2844 uint32_t maxmtu = 0;
2846 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2848 if (inc->inc_faddr.s_addr != INADDR_ANY) {
2850 if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr,
2851 NHR_REF, 0, &nh4) != 0)
2855 maxmtu = nh4.nh_mtu;
2857 /* Report additional interface capabilities. */
2859 if (ifp->if_capenable & IFCAP_TSO4 &&
2860 ifp->if_hwassist & CSUM_TSO) {
2861 cap->ifcap |= CSUM_TSO;
2862 cap->tsomax = ifp->if_hw_tsomax;
2863 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2864 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2867 fib4_free_nh_ext(inc->inc_fibnum, &nh4);
2875 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2877 struct nhop6_extended nh6;
2878 struct in6_addr dst6;
2881 uint32_t maxmtu = 0;
2883 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2885 if (inc->inc_flags & INC_IPV6MINMTU)
2888 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2889 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
2890 if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0,
2895 maxmtu = nh6.nh_mtu;
2897 /* Report additional interface capabilities. */
2899 if (ifp->if_capenable & IFCAP_TSO6 &&
2900 ifp->if_hwassist & CSUM_TSO) {
2901 cap->ifcap |= CSUM_TSO;
2902 cap->tsomax = ifp->if_hw_tsomax;
2903 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2904 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2907 fib6_free_nh_ext(inc->inc_fibnum, &nh6);
2915 * Calculate effective SMSS per RFC5681 definition for a given TCP
2916 * connection at its current state, taking into account SACK and etc.
2919 tcp_maxseg(const struct tcpcb *tp)
2923 if (tp->t_flags & TF_NOOPT)
2924 return (tp->t_maxseg);
2927 * Here we have a simplified code from tcp_addoptions(),
2928 * without a proper loop, and having most of paddings hardcoded.
2929 * We might make mistakes with padding here in some edge cases,
2930 * but this is harmless, since result of tcp_maxseg() is used
2931 * only in cwnd and ssthresh estimations.
2933 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4)
2934 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2935 if (tp->t_flags & TF_RCVD_TSTMP)
2936 optlen = TCPOLEN_TSTAMP_APPA;
2939 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2940 if (tp->t_flags & TF_SIGNATURE)
2941 optlen += PAD(TCPOLEN_SIGNATURE);
2943 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
2944 optlen += TCPOLEN_SACKHDR;
2945 optlen += tp->rcv_numsacks * TCPOLEN_SACK;
2946 optlen = PAD(optlen);
2949 if (tp->t_flags & TF_REQ_TSTMP)
2950 optlen = TCPOLEN_TSTAMP_APPA;
2952 optlen = PAD(TCPOLEN_MAXSEG);
2953 if (tp->t_flags & TF_REQ_SCALE)
2954 optlen += PAD(TCPOLEN_WINDOW);
2955 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2956 if (tp->t_flags & TF_SIGNATURE)
2957 optlen += PAD(TCPOLEN_SIGNATURE);
2959 if (tp->t_flags & TF_SACK_PERMIT)
2960 optlen += PAD(TCPOLEN_SACK_PERMITTED);
2963 optlen = min(optlen, TCP_MAXOLEN);
2964 return (tp->t_maxseg - optlen);
2968 sysctl_drop(SYSCTL_HANDLER_ARGS)
2970 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2971 struct sockaddr_storage addrs[2];
2975 struct sockaddr_in *fin, *lin;
2976 struct epoch_tracker et;
2978 struct sockaddr_in6 *fin6, *lin6;
2989 if (req->oldptr != NULL || req->oldlen != 0)
2991 if (req->newptr == NULL)
2993 if (req->newlen < sizeof(addrs))
2995 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2999 switch (addrs[0].ss_family) {
3002 fin6 = (struct sockaddr_in6 *)&addrs[0];
3003 lin6 = (struct sockaddr_in6 *)&addrs[1];
3004 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3005 lin6->sin6_len != sizeof(struct sockaddr_in6))
3007 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3008 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3010 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3011 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3012 fin = (struct sockaddr_in *)&addrs[0];
3013 lin = (struct sockaddr_in *)&addrs[1];
3016 error = sa6_embedscope(fin6, V_ip6_use_defzone);
3019 error = sa6_embedscope(lin6, V_ip6_use_defzone);
3026 fin = (struct sockaddr_in *)&addrs[0];
3027 lin = (struct sockaddr_in *)&addrs[1];
3028 if (fin->sin_len != sizeof(struct sockaddr_in) ||
3029 lin->sin_len != sizeof(struct sockaddr_in))
3036 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
3037 switch (addrs[0].ss_family) {
3040 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3041 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3042 INPLOOKUP_WLOCKPCB, NULL);
3047 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3048 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3053 if (inp->inp_flags & INP_TIMEWAIT) {
3055 * XXXRW: There currently exists a state where an
3056 * inpcb is present, but its timewait state has been
3057 * discarded. For now, don't allow dropping of this
3065 } else if (!(inp->inp_flags & INP_DROPPED) &&
3066 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
3067 tp = intotcpcb(inp);
3068 tp = tcp_drop(tp, ECONNABORTED);
3075 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
3079 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
3080 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
3081 0, sysctl_drop, "", "Drop TCP connection");
3085 sysctl_switch_tls(SYSCTL_HANDLER_ARGS)
3087 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
3088 struct sockaddr_storage addrs[2];
3090 struct sockaddr_in *fin, *lin;
3091 struct epoch_tracker et;
3093 struct sockaddr_in6 *fin6, *lin6;
3104 if (req->oldptr != NULL || req->oldlen != 0)
3106 if (req->newptr == NULL)
3108 if (req->newlen < sizeof(addrs))
3110 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3114 switch (addrs[0].ss_family) {
3117 fin6 = (struct sockaddr_in6 *)&addrs[0];
3118 lin6 = (struct sockaddr_in6 *)&addrs[1];
3119 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3120 lin6->sin6_len != sizeof(struct sockaddr_in6))
3122 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3123 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3125 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3126 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3127 fin = (struct sockaddr_in *)&addrs[0];
3128 lin = (struct sockaddr_in *)&addrs[1];
3131 error = sa6_embedscope(fin6, V_ip6_use_defzone);
3134 error = sa6_embedscope(lin6, V_ip6_use_defzone);
3141 fin = (struct sockaddr_in *)&addrs[0];
3142 lin = (struct sockaddr_in *)&addrs[1];
3143 if (fin->sin_len != sizeof(struct sockaddr_in) ||
3144 lin->sin_len != sizeof(struct sockaddr_in))
3151 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
3152 switch (addrs[0].ss_family) {
3155 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3156 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3157 INPLOOKUP_WLOCKPCB, NULL);
3162 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3163 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3167 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
3169 if ((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) != 0 ||
3170 inp->inp_socket == NULL) {
3176 so = inp->inp_socket;
3178 error = ktls_set_tx_mode(so,
3179 arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET);
3189 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls,
3190 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
3191 0, sysctl_switch_tls, "", "Switch TCP connection to SW TLS");
3192 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls,
3193 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
3194 1, sysctl_switch_tls, "", "Switch TCP connection to ifnet TLS");
3198 * Generate a standardized TCP log line for use throughout the
3199 * tcp subsystem. Memory allocation is done with M_NOWAIT to
3200 * allow use in the interrupt context.
3202 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
3203 * NB: The function may return NULL if memory allocation failed.
3205 * Due to header inclusion and ordering limitations the struct ip
3206 * and ip6_hdr pointers have to be passed as void pointers.
3209 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3213 /* Is logging enabled? */
3214 if (tcp_log_in_vain == 0)
3217 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3221 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3225 /* Is logging enabled? */
3226 if (tcp_log_debug == 0)
3229 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3233 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3240 const struct ip6_hdr *ip6;
3242 ip6 = (const struct ip6_hdr *)ip6hdr;
3244 ip = (struct ip *)ip4hdr;
3247 * The log line looks like this:
3248 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
3250 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
3251 sizeof(PRINT_TH_FLAGS) + 1 +
3253 2 * INET6_ADDRSTRLEN;
3255 2 * INET_ADDRSTRLEN;
3258 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
3262 strcat(s, "TCP: [");
3265 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
3266 inet_ntoa_r(inc->inc_faddr, sp);
3268 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3270 inet_ntoa_r(inc->inc_laddr, sp);
3272 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3275 ip6_sprintf(sp, &inc->inc6_faddr);
3277 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3279 ip6_sprintf(sp, &inc->inc6_laddr);
3281 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3282 } else if (ip6 && th) {
3283 ip6_sprintf(sp, &ip6->ip6_src);
3285 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3287 ip6_sprintf(sp, &ip6->ip6_dst);
3289 sprintf(sp, "]:%i", ntohs(th->th_dport));
3292 } else if (ip && th) {
3293 inet_ntoa_r(ip->ip_src, sp);
3295 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3297 inet_ntoa_r(ip->ip_dst, sp);
3299 sprintf(sp, "]:%i", ntohs(th->th_dport));
3307 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
3308 if (*(s + size - 1) != '\0')
3309 panic("%s: string too long", __func__);
3314 * A subroutine which makes it easy to track TCP state changes with DTrace.
3315 * This function shouldn't be called for t_state initializations that don't
3316 * correspond to actual TCP state transitions.
3319 tcp_state_change(struct tcpcb *tp, int newstate)
3321 #if defined(KDTRACE_HOOKS)
3322 int pstate = tp->t_state;
3325 TCPSTATES_DEC(tp->t_state);
3326 TCPSTATES_INC(newstate);
3327 tp->t_state = newstate;
3328 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
3332 * Create an external-format (``xtcpcb'') structure using the information in
3333 * the kernel-format tcpcb structure pointed to by tp. This is done to
3334 * reduce the spew of irrelevant information over this interface, to isolate
3335 * user code from changes in the kernel structure, and potentially to provide
3336 * information-hiding if we decide that some of this information should be
3337 * hidden from users.
3340 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
3342 struct tcpcb *tp = intotcpcb(inp);
3345 bzero(xt, sizeof(*xt));
3346 if (inp->inp_flags & INP_TIMEWAIT) {
3347 xt->t_state = TCPS_TIME_WAIT;
3349 xt->t_state = tp->t_state;
3350 xt->t_logstate = tp->t_logstate;
3351 xt->t_flags = tp->t_flags;
3352 xt->t_sndzerowin = tp->t_sndzerowin;
3353 xt->t_sndrexmitpack = tp->t_sndrexmitpack;
3354 xt->t_rcvoopack = tp->t_rcvoopack;
3356 now = getsbinuptime();
3357 #define COPYTIMER(ttt) do { \
3358 if (callout_active(&tp->t_timers->ttt)) \
3359 xt->ttt = (tp->t_timers->ttt.c_time - now) / \
3364 COPYTIMER(tt_delack);
3365 COPYTIMER(tt_rexmt);
3366 COPYTIMER(tt_persist);
3370 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
3372 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
3373 TCP_FUNCTION_NAME_LEN_MAX);
3375 (void)tcp_log_get_id(tp, xt->xt_logid);
3379 xt->xt_len = sizeof(struct xtcpcb);
3380 in_pcbtoxinpcb(inp, &xt->xt_inp);
3381 if (inp->inp_socket == NULL)
3382 xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP;