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 VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1;
199 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW,
200 &VNET_NAME(tcp_ts_offset_per_conn), 0,
201 "Initialize TCP timestamps per connection instead of per host pair");
203 static int tcp_log_debug = 0;
204 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
205 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
207 static int tcp_tcbhashsize;
208 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
209 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
211 static int do_tcpdrain = 1;
212 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
213 "Enable tcp_drain routine for extra help when low on mbufs");
215 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
216 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
218 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1;
219 #define V_icmp_may_rst VNET(icmp_may_rst)
220 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
221 &VNET_NAME(icmp_may_rst), 0,
222 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
224 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0;
225 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
226 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
227 &VNET_NAME(tcp_isn_reseed_interval), 0,
228 "Seconds between reseeding of ISN secret");
230 static int tcp_soreceive_stream;
231 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
232 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
234 VNET_DEFINE(uma_zone_t, sack_hole_zone);
235 #define V_sack_hole_zone VNET(sack_hole_zone)
238 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
241 #define TS_OFFSET_SECRET_LENGTH 32
242 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]);
243 #define V_ts_offset_secret VNET(ts_offset_secret)
245 static int tcp_default_fb_init(struct tcpcb *tp);
246 static void tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged);
247 static int tcp_default_handoff_ok(struct tcpcb *tp);
248 static struct inpcb *tcp_notify(struct inpcb *, int);
249 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
250 static void tcp_mtudisc(struct inpcb *, int);
251 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
252 void *ip4hdr, const void *ip6hdr);
255 static struct tcp_function_block tcp_def_funcblk = {
256 .tfb_tcp_block_name = "freebsd",
257 .tfb_tcp_output = tcp_output,
258 .tfb_tcp_do_segment = tcp_do_segment,
259 .tfb_tcp_ctloutput = tcp_default_ctloutput,
260 .tfb_tcp_handoff_ok = tcp_default_handoff_ok,
261 .tfb_tcp_fb_init = tcp_default_fb_init,
262 .tfb_tcp_fb_fini = tcp_default_fb_fini,
265 int t_functions_inited = 0;
266 static int tcp_fb_cnt = 0;
267 struct tcp_funchead t_functions;
268 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
271 init_tcp_functions(void)
273 if (t_functions_inited == 0) {
274 TAILQ_INIT(&t_functions);
275 rw_init_flags(&tcp_function_lock, "tcp_func_lock" , 0);
276 t_functions_inited = 1;
280 static struct tcp_function_block *
281 find_tcp_functions_locked(struct tcp_function_set *fs)
283 struct tcp_function *f;
284 struct tcp_function_block *blk=NULL;
286 TAILQ_FOREACH(f, &t_functions, tf_next) {
287 if (strcmp(f->tf_name, fs->function_set_name) == 0) {
295 static struct tcp_function_block *
296 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
298 struct tcp_function_block *rblk=NULL;
299 struct tcp_function *f;
301 TAILQ_FOREACH(f, &t_functions, tf_next) {
302 if (f->tf_fb == blk) {
313 struct tcp_function_block *
314 find_and_ref_tcp_functions(struct tcp_function_set *fs)
316 struct tcp_function_block *blk;
318 rw_rlock(&tcp_function_lock);
319 blk = find_tcp_functions_locked(fs);
321 refcount_acquire(&blk->tfb_refcnt);
322 rw_runlock(&tcp_function_lock);
326 struct tcp_function_block *
327 find_and_ref_tcp_fb(struct tcp_function_block *blk)
329 struct tcp_function_block *rblk;
331 rw_rlock(&tcp_function_lock);
332 rblk = find_tcp_fb_locked(blk, NULL);
334 refcount_acquire(&rblk->tfb_refcnt);
335 rw_runlock(&tcp_function_lock);
339 static struct tcp_function_block *
340 find_and_ref_tcp_default_fb(void)
342 struct tcp_function_block *rblk;
344 rw_rlock(&tcp_function_lock);
345 rblk = tcp_func_set_ptr;
346 refcount_acquire(&rblk->tfb_refcnt);
347 rw_runlock(&tcp_function_lock);
352 tcp_switch_back_to_default(struct tcpcb *tp)
354 struct tcp_function_block *tfb;
356 KASSERT(tp->t_fb != &tcp_def_funcblk,
357 ("%s: called by the built-in default stack", __func__));
360 * Release the old stack. This function will either find a new one
363 if (tp->t_fb->tfb_tcp_fb_fini != NULL)
364 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
365 refcount_release(&tp->t_fb->tfb_refcnt);
368 * Now, we'll find a new function block to use.
369 * Start by trying the current user-selected
370 * default, unless this stack is the user-selected
373 tfb = find_and_ref_tcp_default_fb();
374 if (tfb == tp->t_fb) {
375 refcount_release(&tfb->tfb_refcnt);
378 /* Does the stack accept this connection? */
379 if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL &&
380 (*tfb->tfb_tcp_handoff_ok)(tp)) {
381 refcount_release(&tfb->tfb_refcnt);
384 /* Try to use that stack. */
386 /* Initialize the new stack. If it succeeds, we are done. */
388 if (tp->t_fb->tfb_tcp_fb_init == NULL ||
389 (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0)
393 * Initialization failed. Release the reference count on
396 refcount_release(&tfb->tfb_refcnt);
400 * If that wasn't feasible, use the built-in default
401 * stack which is not allowed to reject anyone.
403 tfb = find_and_ref_tcp_fb(&tcp_def_funcblk);
405 /* there always should be a default */
406 panic("Can't refer to tcp_def_funcblk");
408 if (tfb->tfb_tcp_handoff_ok != NULL) {
409 if ((*tfb->tfb_tcp_handoff_ok) (tp)) {
410 /* The default stack cannot say no */
411 panic("Default stack rejects a new session?");
415 if (tp->t_fb->tfb_tcp_fb_init != NULL &&
416 (*tp->t_fb->tfb_tcp_fb_init)(tp)) {
417 /* The default stack cannot fail */
418 panic("Default stack initialization failed");
423 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
426 struct tcp_function_set fs;
427 struct tcp_function_block *blk;
429 memset(&fs, 0, sizeof(fs));
430 rw_rlock(&tcp_function_lock);
431 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
434 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
435 fs.pcbcnt = blk->tfb_refcnt;
437 rw_runlock(&tcp_function_lock);
438 error = sysctl_handle_string(oidp, fs.function_set_name,
439 sizeof(fs.function_set_name), req);
441 /* Check for error or no change */
442 if (error != 0 || req->newptr == NULL)
445 rw_wlock(&tcp_function_lock);
446 blk = find_tcp_functions_locked(&fs);
448 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
452 tcp_func_set_ptr = blk;
454 rw_wunlock(&tcp_function_lock);
458 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
459 CTLTYPE_STRING | CTLFLAG_RW,
460 NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
461 "Set/get the default TCP functions");
464 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
466 int error, cnt, linesz;
467 struct tcp_function *f;
473 rw_rlock(&tcp_function_lock);
474 TAILQ_FOREACH(f, &t_functions, tf_next) {
477 rw_runlock(&tcp_function_lock);
479 bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
480 buffer = malloc(bufsz, M_TEMP, M_WAITOK);
485 linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
486 "Alias", "PCB count");
491 rw_rlock(&tcp_function_lock);
492 TAILQ_FOREACH(f, &t_functions, tf_next) {
493 alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
494 linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
495 f->tf_fb->tfb_tcp_block_name,
496 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
497 alias ? f->tf_name : "-",
498 f->tf_fb->tfb_refcnt);
499 if (linesz >= bufsz) {
507 rw_runlock(&tcp_function_lock);
509 error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
510 free(buffer, M_TEMP);
514 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
515 CTLTYPE_STRING|CTLFLAG_RD,
516 NULL, 0, sysctl_net_inet_list_available, "A",
517 "list available TCP Function sets");
520 * Exports one (struct tcp_function_info) for each alias/name.
523 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS)
526 struct tcp_function *f;
527 struct tcp_function_info tfi;
530 * We don't allow writes.
532 if (req->newptr != NULL)
536 * Wire the old buffer so we can directly copy the functions to
537 * user space without dropping the lock.
539 if (req->oldptr != NULL) {
540 error = sysctl_wire_old_buffer(req, 0);
546 * Walk the list and copy out matching entries. If INVARIANTS
547 * is compiled in, also walk the list to verify the length of
548 * the list matches what we have recorded.
550 rw_rlock(&tcp_function_lock);
554 if (req->oldptr == NULL) {
559 TAILQ_FOREACH(f, &t_functions, tf_next) {
563 if (req->oldptr != NULL) {
564 bzero(&tfi, sizeof(tfi));
565 tfi.tfi_refcnt = f->tf_fb->tfb_refcnt;
566 tfi.tfi_id = f->tf_fb->tfb_id;
567 (void)strncpy(tfi.tfi_alias, f->tf_name,
568 TCP_FUNCTION_NAME_LEN_MAX);
569 tfi.tfi_alias[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
570 (void)strncpy(tfi.tfi_name,
571 f->tf_fb->tfb_tcp_block_name,
572 TCP_FUNCTION_NAME_LEN_MAX);
573 tfi.tfi_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
574 error = SYSCTL_OUT(req, &tfi, sizeof(tfi));
576 * Don't stop on error, as that is the
577 * mechanism we use to accumulate length
578 * information if the buffer was too short.
582 KASSERT(cnt == tcp_fb_cnt,
583 ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt));
587 rw_runlock(&tcp_function_lock);
588 if (req->oldptr == NULL)
589 error = SYSCTL_OUT(req, NULL,
590 (cnt + 1) * sizeof(struct tcp_function_info));
595 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info,
596 CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE,
597 NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info",
598 "List TCP function block name-to-ID mappings");
601 * tfb_tcp_handoff_ok() function for the default stack.
602 * Note that we'll basically try to take all comers.
605 tcp_default_handoff_ok(struct tcpcb *tp)
612 * tfb_tcp_fb_init() function for the default stack.
614 * This handles making sure we have appropriate timers set if you are
615 * transitioning a socket that has some amount of setup done.
617 * The init() fuction from the default can *never* return non-zero i.e.
618 * it is required to always succeed since it is the stack of last resort!
621 tcp_default_fb_init(struct tcpcb *tp)
626 INP_WLOCK_ASSERT(tp->t_inpcb);
628 KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT,
629 ("%s: connection %p in unexpected state %d", __func__, tp,
633 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't
634 * know what to do for unexpected states (which includes TIME_WAIT).
636 if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT)
640 * Make sure some kind of transmission timer is set if there is
643 so = tp->t_inpcb->inp_socket;
644 if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) ||
645 tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) ||
646 tcp_timer_active(tp, TT_PERSIST))) {
648 * If the session has established and it looks like it should
649 * be in the persist state, set the persist timer. Otherwise,
650 * set the retransmit timer.
652 if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 &&
653 (int32_t)(tp->snd_nxt - tp->snd_una) <
654 (int32_t)sbavail(&so->so_snd))
657 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
660 /* All non-embryonic sessions get a keepalive timer. */
661 if (!tcp_timer_active(tp, TT_KEEP))
662 tcp_timer_activate(tp, TT_KEEP,
663 TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) :
670 * tfb_tcp_fb_fini() function for the default stack.
672 * This changes state as necessary (or prudent) to prepare for another stack
673 * to assume responsibility for the connection.
676 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged)
679 INP_WLOCK_ASSERT(tp->t_inpcb);
684 * Target size of TCP PCB hash tables. Must be a power of two.
686 * Note that this can be overridden by the kernel environment
687 * variable net.inet.tcp.tcbhashsize
690 #define TCBHASHSIZE 0
695 * Callouts should be moved into struct tcp directly. They are currently
696 * separate because the tcpcb structure is exported to userland for sysctl
697 * parsing purposes, which do not know about callouts.
708 VNET_DEFINE_STATIC(uma_zone_t, tcpcb_zone);
709 #define V_tcpcb_zone VNET(tcpcb_zone)
711 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
712 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
714 static struct mtx isn_mtx;
716 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
717 #define ISN_LOCK() mtx_lock(&isn_mtx)
718 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
721 * TCP initialization.
724 tcp_zone_change(void *tag)
727 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
728 uma_zone_set_max(V_tcpcb_zone, maxsockets);
729 tcp_tw_zone_change();
733 tcp_inpcb_init(void *mem, int size, int flags)
735 struct inpcb *inp = mem;
737 INP_LOCK_INIT(inp, "inp", "tcpinp");
742 * Take a value and get the next power of 2 that doesn't overflow.
743 * Used to size the tcp_inpcb hash buckets.
746 maketcp_hashsize(int size)
752 * get the next power of 2 higher than maxsockets.
754 hashsize = 1 << fls(size);
755 /* catch overflow, and just go one power of 2 smaller */
756 if (hashsize < size) {
757 hashsize = 1 << (fls(size) - 1);
762 static volatile int next_tcp_stack_id = 1;
765 * Register a TCP function block with the name provided in the names
766 * array. (Note that this function does NOT automatically register
767 * blk->tfb_tcp_block_name as a stack name. Therefore, you should
768 * explicitly include blk->tfb_tcp_block_name in the list of names if
769 * you wish to register the stack with that name.)
771 * Either all name registrations will succeed or all will fail. If
772 * a name registration fails, the function will update the num_names
773 * argument to point to the array index of the name that encountered
776 * Returns 0 on success, or an error code on failure.
779 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
780 const char *names[], int *num_names)
782 struct tcp_function *n;
783 struct tcp_function_set fs;
786 KASSERT(names != NULL && *num_names > 0,
787 ("%s: Called with 0-length name list", __func__));
788 KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
790 if (t_functions_inited == 0) {
791 init_tcp_functions();
793 if ((blk->tfb_tcp_output == NULL) ||
794 (blk->tfb_tcp_do_segment == NULL) ||
795 (blk->tfb_tcp_ctloutput == NULL) ||
796 (strlen(blk->tfb_tcp_block_name) == 0)) {
798 * These functions are required and you
804 if (blk->tfb_tcp_timer_stop_all ||
805 blk->tfb_tcp_timer_activate ||
806 blk->tfb_tcp_timer_active ||
807 blk->tfb_tcp_timer_stop) {
809 * If you define one timer function you
810 * must have them all.
812 if ((blk->tfb_tcp_timer_stop_all == NULL) ||
813 (blk->tfb_tcp_timer_activate == NULL) ||
814 (blk->tfb_tcp_timer_active == NULL) ||
815 (blk->tfb_tcp_timer_stop == NULL)) {
821 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
826 refcount_init(&blk->tfb_refcnt, 0);
827 blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1);
828 for (i = 0; i < *num_names; i++) {
829 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
836 (void)strncpy(fs.function_set_name, names[i],
837 TCP_FUNCTION_NAME_LEN_MAX);
838 fs.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
839 rw_wlock(&tcp_function_lock);
840 if (find_tcp_functions_locked(&fs) != NULL) {
841 /* Duplicate name space not allowed */
842 rw_wunlock(&tcp_function_lock);
843 free(n, M_TCPFUNCTIONS);
847 (void)strncpy(n->tf_name, names[i], TCP_FUNCTION_NAME_LEN_MAX);
848 n->tf_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
849 TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
851 rw_wunlock(&tcp_function_lock);
857 * Deregister the names we just added. Because registration failed
858 * for names[i], we don't need to deregister that name.
861 rw_wlock(&tcp_function_lock);
863 TAILQ_FOREACH(n, &t_functions, tf_next) {
864 if (!strncmp(n->tf_name, names[i],
865 TCP_FUNCTION_NAME_LEN_MAX)) {
866 TAILQ_REMOVE(&t_functions, n, tf_next);
869 free(n, M_TCPFUNCTIONS);
874 rw_wunlock(&tcp_function_lock);
879 * Register a TCP function block using the name provided in the name
882 * Returns 0 on success, or an error code on failure.
885 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
888 const char *name_list[1];
895 name_list[0] = blk->tfb_tcp_block_name;
896 rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
901 * Register a TCP function block using the name defined in
902 * blk->tfb_tcp_block_name.
904 * Returns 0 on success, or an error code on failure.
907 register_tcp_functions(struct tcp_function_block *blk, int wait)
910 return (register_tcp_functions_as_name(blk, NULL, wait));
914 * Deregister all names associated with a function block. This
915 * functionally removes the function block from use within the system.
917 * When called with a true quiesce argument, mark the function block
918 * as being removed so no more stacks will use it and determine
919 * whether the removal would succeed.
921 * When called with a false quiesce argument, actually attempt the
924 * When called with a force argument, attempt to switch all TCBs to
925 * use the default stack instead of returning EBUSY.
927 * Returns 0 on success (or if the removal would succeed, or an error
931 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce,
934 struct tcp_function *f;
936 if (strcmp(blk->tfb_tcp_block_name, "default") == 0) {
937 /* You can't un-register the default */
940 rw_wlock(&tcp_function_lock);
941 if (blk == tcp_func_set_ptr) {
942 /* You can't free the current default */
943 rw_wunlock(&tcp_function_lock);
946 /* Mark the block so no more stacks can use it. */
947 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
949 * If TCBs are still attached to the stack, attempt to switch them
950 * to the default stack.
952 if (force && blk->tfb_refcnt) {
955 VNET_ITERATOR_DECL(vnet_iter);
957 rw_wunlock(&tcp_function_lock);
960 VNET_FOREACH(vnet_iter) {
961 CURVNET_SET(vnet_iter);
962 INP_INFO_WLOCK(&V_tcbinfo);
963 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
965 if (inp->inp_flags & INP_TIMEWAIT) {
970 if (tp == NULL || tp->t_fb != blk) {
974 tcp_switch_back_to_default(tp);
977 INP_INFO_WUNLOCK(&V_tcbinfo);
982 rw_wlock(&tcp_function_lock);
984 if (blk->tfb_refcnt) {
985 /* TCBs still attached. */
986 rw_wunlock(&tcp_function_lock);
991 rw_wunlock(&tcp_function_lock);
994 /* Remove any function names that map to this function block. */
995 while (find_tcp_fb_locked(blk, &f) != NULL) {
996 TAILQ_REMOVE(&t_functions, f, tf_next);
999 free(f, M_TCPFUNCTIONS);
1001 rw_wunlock(&tcp_function_lock);
1008 const char *tcbhash_tuneable;
1011 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
1014 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
1015 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1016 printf("%s: WARNING: unable to register helper hook\n", __func__);
1017 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
1018 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1019 printf("%s: WARNING: unable to register helper hook\n", __func__);
1021 hashsize = TCBHASHSIZE;
1022 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
1023 if (hashsize == 0) {
1025 * Auto tune the hash size based on maxsockets.
1026 * A perfect hash would have a 1:1 mapping
1027 * (hashsize = maxsockets) however it's been
1028 * suggested that O(2) average is better.
1030 hashsize = maketcp_hashsize(maxsockets / 4);
1032 * Our historical default is 512,
1033 * do not autotune lower than this.
1037 if (bootverbose && IS_DEFAULT_VNET(curvnet))
1038 printf("%s: %s auto tuned to %d\n", __func__,
1039 tcbhash_tuneable, hashsize);
1042 * We require a hashsize to be a power of two.
1043 * Previously if it was not a power of two we would just reset it
1044 * back to 512, which could be a nasty surprise if you did not notice
1045 * the error message.
1046 * Instead what we do is clip it to the closest power of two lower
1047 * than the specified hash value.
1049 if (!powerof2(hashsize)) {
1050 int oldhashsize = hashsize;
1052 hashsize = maketcp_hashsize(hashsize);
1053 /* prevent absurdly low value */
1056 printf("%s: WARNING: TCB hash size not a power of 2, "
1057 "clipped from %d to %d.\n", __func__, oldhashsize,
1060 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
1061 "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE);
1064 * These have to be type stable for the benefit of the timers.
1066 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
1067 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1068 uma_zone_set_max(V_tcpcb_zone, maxsockets);
1069 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
1075 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
1076 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
1077 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1079 tcp_fastopen_init();
1081 /* Skip initialization of globals for non-default instances. */
1082 if (!IS_DEFAULT_VNET(curvnet))
1085 tcp_reass_global_init();
1087 /* XXX virtualize those bellow? */
1088 tcp_delacktime = TCPTV_DELACK;
1089 tcp_keepinit = TCPTV_KEEP_INIT;
1090 tcp_keepidle = TCPTV_KEEP_IDLE;
1091 tcp_keepintvl = TCPTV_KEEPINTVL;
1092 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
1093 tcp_msl = TCPTV_MSL;
1094 tcp_rexmit_initial = TCPTV_RTOBASE;
1095 if (tcp_rexmit_initial < 1)
1096 tcp_rexmit_initial = 1;
1097 tcp_rexmit_min = TCPTV_MIN;
1098 if (tcp_rexmit_min < 1)
1100 tcp_persmin = TCPTV_PERSMIN;
1101 tcp_persmax = TCPTV_PERSMAX;
1102 tcp_rexmit_slop = TCPTV_CPU_VAR;
1103 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
1104 tcp_tcbhashsize = hashsize;
1105 /* Setup the tcp function block list */
1106 init_tcp_functions();
1107 register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
1109 /* Initialize the TCP logging data. */
1112 arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0);
1114 if (tcp_soreceive_stream) {
1116 tcp_usrreqs.pru_soreceive = soreceive_stream;
1119 tcp6_usrreqs.pru_soreceive = soreceive_stream;
1124 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
1126 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
1128 if (max_protohdr < TCP_MINPROTOHDR)
1129 max_protohdr = TCP_MINPROTOHDR;
1130 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
1132 #undef TCP_MINPROTOHDR
1135 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
1136 SHUTDOWN_PRI_DEFAULT);
1137 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
1138 EVENTHANDLER_PRI_ANY);
1146 tcp_destroy(void *unused __unused)
1154 * All our processes are gone, all our sockets should be cleaned
1155 * up, which means, we should be past the tcp_discardcb() calls.
1156 * Sleep to let all tcpcb timers really disappear and cleanup.
1159 INP_LIST_RLOCK(&V_tcbinfo);
1160 n = V_tcbinfo.ipi_count;
1161 INP_LIST_RUNLOCK(&V_tcbinfo);
1164 pause("tcpdes", hz / 10);
1169 in_pcbinfo_destroy(&V_tcbinfo);
1170 /* tcp_discardcb() clears the sack_holes up. */
1171 uma_zdestroy(V_sack_hole_zone);
1172 uma_zdestroy(V_tcpcb_zone);
1175 * Cannot free the zone until all tcpcbs are released as we attach
1176 * the allocations to them.
1178 tcp_fastopen_destroy();
1181 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
1183 printf("%s: WARNING: unable to deregister helper hook "
1184 "type=%d, id=%d: error %d returned\n", __func__,
1185 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
1187 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
1189 printf("%s: WARNING: unable to deregister helper hook "
1190 "type=%d, id=%d: error %d returned\n", __func__,
1191 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
1195 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
1205 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
1206 * tcp_template used to store this data in mbufs, but we now recopy it out
1207 * of the tcpcb each time to conserve mbufs.
1210 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
1212 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
1214 INP_WLOCK_ASSERT(inp);
1217 if ((inp->inp_vflag & INP_IPV6) != 0) {
1218 struct ip6_hdr *ip6;
1220 ip6 = (struct ip6_hdr *)ip_ptr;
1221 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
1222 (inp->inp_flow & IPV6_FLOWINFO_MASK);
1223 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
1224 (IPV6_VERSION & IPV6_VERSION_MASK);
1225 ip6->ip6_nxt = IPPROTO_TCP;
1226 ip6->ip6_plen = htons(sizeof(struct tcphdr));
1227 ip6->ip6_src = inp->in6p_laddr;
1228 ip6->ip6_dst = inp->in6p_faddr;
1231 #if defined(INET6) && defined(INET)
1238 ip = (struct ip *)ip_ptr;
1239 ip->ip_v = IPVERSION;
1241 ip->ip_tos = inp->inp_ip_tos;
1245 ip->ip_ttl = inp->inp_ip_ttl;
1247 ip->ip_p = IPPROTO_TCP;
1248 ip->ip_src = inp->inp_laddr;
1249 ip->ip_dst = inp->inp_faddr;
1252 th->th_sport = inp->inp_lport;
1253 th->th_dport = inp->inp_fport;
1261 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
1265 * Create template to be used to send tcp packets on a connection.
1266 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
1267 * use for this function is in keepalives, which use tcp_respond.
1270 tcpip_maketemplate(struct inpcb *inp)
1274 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
1277 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
1282 * Send a single message to the TCP at address specified by
1283 * the given TCP/IP header. If m == NULL, then we make a copy
1284 * of the tcpiphdr at th and send directly to the addressed host.
1285 * This is used to force keep alive messages out using the TCP
1286 * template for a connection. If flags are given then we send
1287 * a message back to the TCP which originated the segment th,
1288 * and discard the mbuf containing it and any other attached mbufs.
1290 * In any case the ack and sequence number of the transmitted
1291 * segment are as specified by the parameters.
1293 * NOTE: If m != NULL, then th must point to *inside* the mbuf.
1296 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
1297 tcp_seq ack, tcp_seq seq, int flags)
1306 struct ip6_hdr *ip6;
1309 int optlen, tlen, win;
1312 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
1315 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
1322 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
1323 INP_WLOCK_ASSERT(inp);
1330 if (!(flags & TH_RST)) {
1331 win = sbspace(&inp->inp_socket->so_rcv);
1332 if (win > TCP_MAXWIN << tp->rcv_scale)
1333 win = TCP_MAXWIN << tp->rcv_scale;
1335 if ((tp->t_flags & TF_NOOPT) == 0)
1339 m = m_gethdr(M_NOWAIT, MT_DATA);
1342 m->m_data += max_linkhdr;
1345 bcopy((caddr_t)ip6, mtod(m, caddr_t),
1346 sizeof(struct ip6_hdr));
1347 ip6 = mtod(m, struct ip6_hdr *);
1348 nth = (struct tcphdr *)(ip6 + 1);
1352 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1353 ip = mtod(m, struct ip *);
1354 nth = (struct tcphdr *)(ip + 1);
1356 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1358 } else if (!M_WRITABLE(m)) {
1361 /* Can't reuse 'm', allocate a new mbuf. */
1362 n = m_gethdr(M_NOWAIT, MT_DATA);
1368 if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1374 n->m_data += max_linkhdr;
1375 /* m_len is set later */
1376 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1379 bcopy((caddr_t)ip6, mtod(n, caddr_t),
1380 sizeof(struct ip6_hdr));
1381 ip6 = mtod(n, struct ip6_hdr *);
1382 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1383 nth = (struct tcphdr *)(ip6 + 1);
1387 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1388 ip = mtod(n, struct ip *);
1389 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1390 nth = (struct tcphdr *)(ip + 1);
1392 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1393 xchg(nth->th_dport, nth->th_sport, uint16_t);
1400 * XXX MRT We inherit the FIB, which is lucky.
1404 m->m_data = (caddr_t)ipgen;
1405 /* m_len is set later */
1408 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1409 nth = (struct tcphdr *)(ip6 + 1);
1413 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1414 nth = (struct tcphdr *)(ip + 1);
1418 * this is usually a case when an extension header
1419 * exists between the IPv6 header and the
1422 nth->th_sport = th->th_sport;
1423 nth->th_dport = th->th_dport;
1425 xchg(nth->th_dport, nth->th_sport, uint16_t);
1431 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1433 #if defined(INET) && defined(INET6)
1437 tlen = sizeof (struct tcpiphdr);
1441 KASSERT(M_TRAILINGSPACE(m) >= tlen,
1442 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1443 m, tlen, (long)M_TRAILINGSPACE(m)));
1448 /* Make sure we have room. */
1449 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1450 m->m_next = m_get(M_NOWAIT, MT_DATA);
1452 optp = mtod(m->m_next, u_char *);
1457 optp = (u_char *) (nth + 1);
1463 if (tp->t_flags & TF_RCVD_TSTMP) {
1464 to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1465 to.to_tsecr = tp->ts_recent;
1466 to.to_flags |= TOF_TS;
1468 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1469 /* TCP-MD5 (RFC2385). */
1470 if (tp->t_flags & TF_SIGNATURE)
1471 to.to_flags |= TOF_SIGNATURE;
1473 /* Add the options. */
1474 tlen += optlen = tcp_addoptions(&to, optp);
1476 /* Update m_len in the correct mbuf. */
1477 optm->m_len += optlen;
1483 ip6->ip6_vfc = IPV6_VERSION;
1484 ip6->ip6_nxt = IPPROTO_TCP;
1485 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1488 #if defined(INET) && defined(INET6)
1493 ip->ip_len = htons(tlen);
1494 ip->ip_ttl = V_ip_defttl;
1495 if (V_path_mtu_discovery)
1496 ip->ip_off |= htons(IP_DF);
1499 m->m_pkthdr.len = tlen;
1500 m->m_pkthdr.rcvif = NULL;
1504 * Packet is associated with a socket, so allow the
1505 * label of the response to reflect the socket label.
1507 INP_WLOCK_ASSERT(inp);
1508 mac_inpcb_create_mbuf(inp, m);
1511 * Packet is not associated with a socket, so possibly
1512 * update the label in place.
1514 mac_netinet_tcp_reply(m);
1517 nth->th_seq = htonl(seq);
1518 nth->th_ack = htonl(ack);
1520 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1521 nth->th_flags = flags;
1523 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1525 nth->th_win = htons((u_short)win);
1528 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1529 if (to.to_flags & TOF_SIGNATURE) {
1530 if (!TCPMD5_ENABLED() ||
1531 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
1538 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1541 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1542 nth->th_sum = in6_cksum_pseudo(ip6,
1543 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
1544 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
1548 #if defined(INET6) && defined(INET)
1553 m->m_pkthdr.csum_flags = CSUM_TCP;
1554 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1555 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1559 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1560 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1562 TCP_PROBE3(debug__output, tp, th, m);
1564 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
1568 TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
1569 (void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
1572 #if defined(INET) && defined(INET6)
1577 TCP_PROBE5(send, NULL, tp, ip, tp, nth);
1578 (void)ip_output(m, NULL, NULL, 0, NULL, inp);
1584 * Create a new TCP control block, making an
1585 * empty reassembly queue and hooking it to the argument
1586 * protocol control block. The `inp' parameter must have
1587 * come from the zone allocator set up in tcp_init().
1590 tcp_newtcpcb(struct inpcb *inp)
1592 struct tcpcb_mem *tm;
1595 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1598 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1603 /* Initialise cc_var struct for this tcpcb. */
1605 tp->ccv->type = IPPROTO_TCP;
1606 tp->ccv->ccvc.tcp = tp;
1607 rw_rlock(&tcp_function_lock);
1608 tp->t_fb = tcp_func_set_ptr;
1609 refcount_acquire(&tp->t_fb->tfb_refcnt);
1610 rw_runlock(&tcp_function_lock);
1612 * Use the current system default CC algorithm.
1615 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1616 CC_ALGO(tp) = CC_DEFAULT();
1619 if (CC_ALGO(tp)->cb_init != NULL)
1620 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1621 if (tp->t_fb->tfb_tcp_fb_fini)
1622 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1623 refcount_release(&tp->t_fb->tfb_refcnt);
1624 uma_zfree(V_tcpcb_zone, tm);
1630 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1631 if (tp->t_fb->tfb_tcp_fb_fini)
1632 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1633 refcount_release(&tp->t_fb->tfb_refcnt);
1634 uma_zfree(V_tcpcb_zone, tm);
1640 tp->t_vnet = inp->inp_vnet;
1642 tp->t_timers = &tm->tt;
1643 TAILQ_INIT(&tp->t_segq);
1646 isipv6 ? V_tcp_v6mssdflt :
1650 /* Set up our timeouts. */
1651 callout_init(&tp->t_timers->tt_rexmt, 1);
1652 callout_init(&tp->t_timers->tt_persist, 1);
1653 callout_init(&tp->t_timers->tt_keep, 1);
1654 callout_init(&tp->t_timers->tt_2msl, 1);
1655 callout_init(&tp->t_timers->tt_delack, 1);
1657 if (V_tcp_do_rfc1323)
1658 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1660 tp->t_flags |= TF_SACK_PERMIT;
1661 TAILQ_INIT(&tp->snd_holes);
1663 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1666 in_pcbref(inp); /* Reference for tcpcb */
1670 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1671 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
1672 * reasonable initial retransmit time.
1674 tp->t_srtt = TCPTV_SRTTBASE;
1675 tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1676 tp->t_rttmin = tcp_rexmit_min;
1677 tp->t_rxtcur = tcp_rexmit_initial;
1678 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1679 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1680 tp->t_rcvtime = ticks;
1682 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1683 * because the socket may be bound to an IPv6 wildcard address,
1684 * which may match an IPv4-mapped IPv6 address.
1686 inp->inp_ip_ttl = V_ip_defttl;
1690 * Init the TCP PCAP queues.
1692 tcp_pcap_tcpcb_init(tp);
1695 /* Initialize the per-TCPCB log data. */
1696 tcp_log_tcpcbinit(tp);
1698 if (tp->t_fb->tfb_tcp_fb_init) {
1699 (*tp->t_fb->tfb_tcp_fb_init)(tp);
1701 return (tp); /* XXX */
1705 * Switch the congestion control algorithm back to NewReno for any active
1706 * control blocks using an algorithm which is about to go away.
1707 * This ensures the CC framework can allow the unload to proceed without leaving
1708 * any dangling pointers which would trigger a panic.
1709 * Returning non-zero would inform the CC framework that something went wrong
1710 * and it would be unsafe to allow the unload to proceed. However, there is no
1711 * way for this to occur with this implementation so we always return zero.
1714 tcp_ccalgounload(struct cc_algo *unload_algo)
1716 struct cc_algo *tmpalgo;
1719 VNET_ITERATOR_DECL(vnet_iter);
1722 * Check all active control blocks across all network stacks and change
1723 * any that are using "unload_algo" back to NewReno. If "unload_algo"
1724 * requires cleanup code to be run, call it.
1727 VNET_FOREACH(vnet_iter) {
1728 CURVNET_SET(vnet_iter);
1729 INP_INFO_WLOCK(&V_tcbinfo);
1731 * New connections already part way through being initialised
1732 * with the CC algo we're removing will not race with this code
1733 * because the INP_INFO_WLOCK is held during initialisation. We
1734 * therefore don't enter the loop below until the connection
1735 * list has stabilised.
1737 CK_LIST_FOREACH(inp, &V_tcb, inp_list) {
1739 /* Important to skip tcptw structs. */
1740 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1741 (tp = intotcpcb(inp)) != NULL) {
1743 * By holding INP_WLOCK here, we are assured
1744 * that the connection is not currently
1745 * executing inside the CC module's functions
1746 * i.e. it is safe to make the switch back to
1749 if (CC_ALGO(tp) == unload_algo) {
1750 tmpalgo = CC_ALGO(tp);
1751 if (tmpalgo->cb_destroy != NULL)
1752 tmpalgo->cb_destroy(tp->ccv);
1755 * NewReno may allocate memory on
1756 * demand for certain stateful
1757 * configuration as needed, but is
1758 * coded to never fail on memory
1759 * allocation failure so it is a safe
1762 CC_ALGO(tp) = &newreno_cc_algo;
1767 INP_INFO_WUNLOCK(&V_tcbinfo);
1770 VNET_LIST_RUNLOCK();
1776 * Drop a TCP connection, reporting
1777 * the specified error. If connection is synchronized,
1778 * then send a RST to peer.
1781 tcp_drop(struct tcpcb *tp, int errno)
1783 struct socket *so = tp->t_inpcb->inp_socket;
1785 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1786 INP_WLOCK_ASSERT(tp->t_inpcb);
1788 if (TCPS_HAVERCVDSYN(tp->t_state)) {
1789 tcp_state_change(tp, TCPS_CLOSED);
1790 (void) tp->t_fb->tfb_tcp_output(tp);
1791 TCPSTAT_INC(tcps_drops);
1793 TCPSTAT_INC(tcps_conndrops);
1794 if (errno == ETIMEDOUT && tp->t_softerror)
1795 errno = tp->t_softerror;
1796 so->so_error = errno;
1797 return (tcp_close(tp));
1801 tcp_discardcb(struct tcpcb *tp)
1803 struct inpcb *inp = tp->t_inpcb;
1804 struct socket *so = inp->inp_socket;
1806 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1808 int released __unused;
1810 INP_WLOCK_ASSERT(inp);
1813 * Make sure that all of our timers are stopped before we delete the
1816 * If stopping a timer fails, we schedule a discard function in same
1817 * callout, and the last discard function called will take care of
1818 * deleting the tcpcb.
1820 tp->t_timers->tt_draincnt = 0;
1821 tcp_timer_stop(tp, TT_REXMT);
1822 tcp_timer_stop(tp, TT_PERSIST);
1823 tcp_timer_stop(tp, TT_KEEP);
1824 tcp_timer_stop(tp, TT_2MSL);
1825 tcp_timer_stop(tp, TT_DELACK);
1826 if (tp->t_fb->tfb_tcp_timer_stop_all) {
1828 * Call the stop-all function of the methods,
1829 * this function should call the tcp_timer_stop()
1830 * method with each of the function specific timeouts.
1831 * That stop will be called via the tfb_tcp_timer_stop()
1832 * which should use the async drain function of the
1833 * callout system (see tcp_var.h).
1835 tp->t_fb->tfb_tcp_timer_stop_all(tp);
1839 * If we got enough samples through the srtt filter,
1840 * save the rtt and rttvar in the routing entry.
1841 * 'Enough' is arbitrarily defined as 4 rtt samples.
1842 * 4 samples is enough for the srtt filter to converge
1843 * to within enough % of the correct value; fewer samples
1844 * and we could save a bogus rtt. The danger is not high
1845 * as tcp quickly recovers from everything.
1846 * XXX: Works very well but needs some more statistics!
1848 if (tp->t_rttupdated >= 4) {
1849 struct hc_metrics_lite metrics;
1852 bzero(&metrics, sizeof(metrics));
1854 * Update the ssthresh always when the conditions below
1855 * are satisfied. This gives us better new start value
1856 * for the congestion avoidance for new connections.
1857 * ssthresh is only set if packet loss occurred on a session.
1859 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1860 * being torn down. Ideally this code would not use 'so'.
1862 ssthresh = tp->snd_ssthresh;
1863 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1865 * convert the limit from user data bytes to
1866 * packets then to packet data bytes.
1868 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1871 ssthresh *= (tp->t_maxseg +
1873 (isipv6 ? sizeof (struct ip6_hdr) +
1874 sizeof (struct tcphdr) :
1876 sizeof (struct tcpiphdr)
1883 metrics.rmx_ssthresh = ssthresh;
1885 metrics.rmx_rtt = tp->t_srtt;
1886 metrics.rmx_rttvar = tp->t_rttvar;
1887 metrics.rmx_cwnd = tp->snd_cwnd;
1888 metrics.rmx_sendpipe = 0;
1889 metrics.rmx_recvpipe = 0;
1891 tcp_hc_update(&inp->inp_inc, &metrics);
1894 /* free the reassembly queue, if any */
1895 tcp_reass_flush(tp);
1898 /* Disconnect offload device, if any. */
1899 if (tp->t_flags & TF_TOE)
1900 tcp_offload_detach(tp);
1903 tcp_free_sackholes(tp);
1906 /* Free the TCP PCAP queues. */
1907 tcp_pcap_drain(&(tp->t_inpkts));
1908 tcp_pcap_drain(&(tp->t_outpkts));
1911 /* Allow the CC algorithm to clean up after itself. */
1912 if (CC_ALGO(tp)->cb_destroy != NULL)
1913 CC_ALGO(tp)->cb_destroy(tp->ccv);
1917 khelp_destroy_osd(tp->osd);
1921 inp->inp_ppcb = NULL;
1922 if (tp->t_timers->tt_draincnt == 0) {
1923 /* We own the last reference on tcpcb, let's free it. */
1925 tcp_log_tcpcbfini(tp);
1927 TCPSTATES_DEC(tp->t_state);
1928 if (tp->t_fb->tfb_tcp_fb_fini)
1929 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1930 refcount_release(&tp->t_fb->tfb_refcnt);
1932 uma_zfree(V_tcpcb_zone, tp);
1933 released = in_pcbrele_wlocked(inp);
1934 KASSERT(!released, ("%s: inp %p should not have been released "
1935 "here", __func__, inp));
1940 tcp_timer_discard(void *ptp)
1944 struct epoch_tracker et;
1946 tp = (struct tcpcb *)ptp;
1947 CURVNET_SET(tp->t_vnet);
1948 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
1950 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1953 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1954 ("%s: tcpcb has to be stopped here", __func__));
1955 tp->t_timers->tt_draincnt--;
1956 if (tp->t_timers->tt_draincnt == 0) {
1957 /* We own the last reference on this tcpcb, let's free it. */
1959 tcp_log_tcpcbfini(tp);
1961 TCPSTATES_DEC(tp->t_state);
1962 if (tp->t_fb->tfb_tcp_fb_fini)
1963 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1964 refcount_release(&tp->t_fb->tfb_refcnt);
1966 uma_zfree(V_tcpcb_zone, tp);
1967 if (in_pcbrele_wlocked(inp)) {
1968 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1974 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1979 * Attempt to close a TCP control block, marking it as dropped, and freeing
1980 * the socket if we hold the only reference.
1983 tcp_close(struct tcpcb *tp)
1985 struct inpcb *inp = tp->t_inpcb;
1988 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1989 INP_WLOCK_ASSERT(inp);
1992 if (tp->t_state == TCPS_LISTEN)
1993 tcp_offload_listen_stop(tp);
1996 * This releases the TFO pending counter resource for TFO listen
1997 * sockets as well as passively-created TFO sockets that transition
1998 * from SYN_RECEIVED to CLOSED.
2000 if (tp->t_tfo_pending) {
2001 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2002 tp->t_tfo_pending = NULL;
2005 TCPSTAT_INC(tcps_closed);
2006 if (tp->t_state != TCPS_CLOSED)
2007 tcp_state_change(tp, TCPS_CLOSED);
2008 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
2009 so = inp->inp_socket;
2010 soisdisconnected(so);
2011 if (inp->inp_flags & INP_SOCKREF) {
2012 KASSERT(so->so_state & SS_PROTOREF,
2013 ("tcp_close: !SS_PROTOREF"));
2014 inp->inp_flags &= ~INP_SOCKREF;
2017 so->so_state &= ~SS_PROTOREF;
2027 VNET_ITERATOR_DECL(vnet_iter);
2032 VNET_LIST_RLOCK_NOSLEEP();
2033 VNET_FOREACH(vnet_iter) {
2034 CURVNET_SET(vnet_iter);
2039 * Walk the tcpbs, if existing, and flush the reassembly queue,
2040 * if there is one...
2041 * XXX: The "Net/3" implementation doesn't imply that the TCP
2042 * reassembly queue should be flushed, but in a situation
2043 * where we're really low on mbufs, this is potentially
2046 INP_INFO_WLOCK(&V_tcbinfo);
2047 CK_LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
2049 if (inpb->inp_flags & INP_TIMEWAIT) {
2053 if ((tcpb = intotcpcb(inpb)) != NULL) {
2054 tcp_reass_flush(tcpb);
2055 tcp_clean_sackreport(tcpb);
2057 tcp_log_drain(tcpb);
2060 if (tcp_pcap_aggressive_free) {
2061 /* Free the TCP PCAP queues. */
2062 tcp_pcap_drain(&(tcpb->t_inpkts));
2063 tcp_pcap_drain(&(tcpb->t_outpkts));
2069 INP_INFO_WUNLOCK(&V_tcbinfo);
2072 VNET_LIST_RUNLOCK_NOSLEEP();
2076 * Notify a tcp user of an asynchronous error;
2077 * store error as soft error, but wake up user
2078 * (for now, won't do anything until can select for soft error).
2080 * Do not wake up user since there currently is no mechanism for
2081 * reporting soft errors (yet - a kqueue filter may be added).
2083 static struct inpcb *
2084 tcp_notify(struct inpcb *inp, int error)
2088 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2089 INP_WLOCK_ASSERT(inp);
2091 if ((inp->inp_flags & INP_TIMEWAIT) ||
2092 (inp->inp_flags & INP_DROPPED))
2095 tp = intotcpcb(inp);
2096 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
2099 * Ignore some errors if we are hooked up.
2100 * If connection hasn't completed, has retransmitted several times,
2101 * and receives a second error, give up now. This is better
2102 * than waiting a long time to establish a connection that
2103 * can never complete.
2105 if (tp->t_state == TCPS_ESTABLISHED &&
2106 (error == EHOSTUNREACH || error == ENETUNREACH ||
2107 error == EHOSTDOWN)) {
2108 if (inp->inp_route.ro_rt) {
2109 RTFREE(inp->inp_route.ro_rt);
2110 inp->inp_route.ro_rt = (struct rtentry *)NULL;
2113 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
2115 tp = tcp_drop(tp, error);
2121 tp->t_softerror = error;
2125 wakeup( &so->so_timeo);
2132 tcp_pcblist(SYSCTL_HANDLER_ARGS)
2134 int error, i, m, n, pcb_count;
2135 struct inpcb *inp, **inp_list;
2138 struct epoch_tracker et;
2141 * The process of preparing the TCB list is too time-consuming and
2142 * resource-intensive to repeat twice on every request.
2144 if (req->oldptr == NULL) {
2145 n = V_tcbinfo.ipi_count +
2146 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2147 n += imax(n / 8, 10);
2148 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
2152 if (req->newptr != NULL)
2156 * OK, now we're committed to doing something.
2158 INP_LIST_RLOCK(&V_tcbinfo);
2159 gencnt = V_tcbinfo.ipi_gencnt;
2160 n = V_tcbinfo.ipi_count;
2161 INP_LIST_RUNLOCK(&V_tcbinfo);
2163 m = counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2165 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
2166 + (n + m) * sizeof(struct xtcpcb));
2170 bzero(&xig, sizeof(xig));
2171 xig.xig_len = sizeof xig;
2172 xig.xig_count = n + m;
2173 xig.xig_gen = gencnt;
2174 xig.xig_sogen = so_gencnt;
2175 error = SYSCTL_OUT(req, &xig, sizeof xig);
2179 error = syncache_pcblist(req, m, &pcb_count);
2183 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
2185 INP_INFO_WLOCK(&V_tcbinfo);
2186 for (inp = CK_LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
2187 inp != NULL && i < n; inp = CK_LIST_NEXT(inp, inp_list)) {
2189 if (inp->inp_gencnt <= gencnt) {
2191 * XXX: This use of cr_cansee(), introduced with
2192 * TCP state changes, is not quite right, but for
2193 * now, better than nothing.
2195 if (inp->inp_flags & INP_TIMEWAIT) {
2196 if (intotw(inp) != NULL)
2197 error = cr_cansee(req->td->td_ucred,
2198 intotw(inp)->tw_cred);
2200 error = EINVAL; /* Skip this inp. */
2202 error = cr_canseeinpcb(req->td->td_ucred, inp);
2205 inp_list[i++] = inp;
2210 INP_INFO_WUNLOCK(&V_tcbinfo);
2214 for (i = 0; i < n; i++) {
2217 if (inp->inp_gencnt <= gencnt) {
2220 tcp_inptoxtp(inp, &xt);
2222 error = SYSCTL_OUT(req, &xt, sizeof xt);
2226 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
2227 for (i = 0; i < n; i++) {
2230 if (!in_pcbrele_rlocked(inp))
2233 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2237 * Give the user an updated idea of our state.
2238 * If the generation differs from what we told
2239 * her before, she knows that something happened
2240 * while we were processing this request, and it
2241 * might be necessary to retry.
2243 INP_LIST_RLOCK(&V_tcbinfo);
2244 xig.xig_gen = V_tcbinfo.ipi_gencnt;
2245 xig.xig_sogen = so_gencnt;
2246 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
2247 INP_LIST_RUNLOCK(&V_tcbinfo);
2248 error = SYSCTL_OUT(req, &xig, sizeof xig);
2250 free(inp_list, M_TEMP);
2254 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
2255 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
2256 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
2260 tcp_getcred(SYSCTL_HANDLER_ARGS)
2263 struct sockaddr_in addrs[2];
2267 error = priv_check(req->td, PRIV_NETINET_GETCRED);
2270 error = SYSCTL_IN(req, addrs, sizeof(addrs));
2273 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
2274 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
2276 if (inp->inp_socket == NULL)
2279 error = cr_canseeinpcb(req->td->td_ucred, inp);
2281 cru2x(inp->inp_cred, &xuc);
2286 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2290 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
2291 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
2292 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
2297 tcp6_getcred(SYSCTL_HANDLER_ARGS)
2300 struct sockaddr_in6 addrs[2];
2307 error = priv_check(req->td, PRIV_NETINET_GETCRED);
2310 error = SYSCTL_IN(req, addrs, sizeof(addrs));
2313 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
2314 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
2317 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
2319 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
2328 inp = in_pcblookup(&V_tcbinfo,
2329 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
2331 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
2332 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
2335 inp = in6_pcblookup(&V_tcbinfo,
2336 &addrs[1].sin6_addr, addrs[1].sin6_port,
2337 &addrs[0].sin6_addr, addrs[0].sin6_port,
2338 INPLOOKUP_RLOCKPCB, NULL);
2340 if (inp->inp_socket == NULL)
2343 error = cr_canseeinpcb(req->td->td_ucred, inp);
2345 cru2x(inp->inp_cred, &xuc);
2350 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2354 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
2355 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
2356 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
2362 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
2364 struct ip *ip = vip;
2366 struct in_addr faddr;
2369 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2371 struct in_conninfo inc;
2372 struct epoch_tracker et;
2373 tcp_seq icmp_tcp_seq;
2376 faddr = ((struct sockaddr_in *)sa)->sin_addr;
2377 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
2380 if (cmd == PRC_MSGSIZE)
2381 notify = tcp_mtudisc_notify;
2382 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2383 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2384 cmd == PRC_TIMXCEED_INTRANS) && ip)
2385 notify = tcp_drop_syn_sent;
2388 * Hostdead is ugly because it goes linearly through all PCBs.
2389 * XXX: We never get this from ICMP, otherwise it makes an
2390 * excellent DoS attack on machines with many connections.
2392 else if (cmd == PRC_HOSTDEAD)
2394 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
2398 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
2402 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
2403 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2404 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
2405 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
2406 th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2407 if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2408 /* signal EHOSTDOWN, as it flushes the cached route */
2409 inp = (*notify)(inp, EHOSTDOWN);
2412 icmp_tcp_seq = th->th_seq;
2414 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2415 !(inp->inp_flags & INP_DROPPED) &&
2416 !(inp->inp_socket == NULL)) {
2417 tp = intotcpcb(inp);
2418 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2419 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2420 if (cmd == PRC_MSGSIZE) {
2423 * If we got a needfrag set the MTU
2424 * in the route to the suggested new
2425 * value (if given) and then notify.
2427 mtu = ntohs(icp->icmp_nextmtu);
2429 * If no alternative MTU was
2430 * proposed, try the next smaller
2435 ntohs(ip->ip_len), 1);
2436 if (mtu < V_tcp_minmss +
2437 sizeof(struct tcpiphdr))
2438 mtu = V_tcp_minmss +
2439 sizeof(struct tcpiphdr);
2441 * Only process the offered MTU if it
2442 * is smaller than the current one.
2444 if (mtu < tp->t_maxseg +
2445 sizeof(struct tcpiphdr)) {
2446 bzero(&inc, sizeof(inc));
2447 inc.inc_faddr = faddr;
2449 inp->inp_inc.inc_fibnum;
2450 tcp_hc_updatemtu(&inc, mtu);
2451 tcp_mtudisc(inp, mtu);
2454 inp = (*notify)(inp,
2455 inetctlerrmap[cmd]);
2459 bzero(&inc, sizeof(inc));
2460 inc.inc_fport = th->th_dport;
2461 inc.inc_lport = th->th_sport;
2462 inc.inc_faddr = faddr;
2463 inc.inc_laddr = ip->ip_src;
2464 syncache_unreach(&inc, icmp_tcp_seq);
2469 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2475 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
2477 struct in6_addr *dst;
2478 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2479 struct ip6_hdr *ip6;
2483 struct icmp6_hdr *icmp6;
2484 struct ip6ctlparam *ip6cp = NULL;
2485 const struct sockaddr_in6 *sa6_src = NULL;
2486 struct in_conninfo inc;
2487 struct epoch_tracker et;
2492 tcp_seq icmp_tcp_seq;
2496 if (sa->sa_family != AF_INET6 ||
2497 sa->sa_len != sizeof(struct sockaddr_in6))
2500 /* if the parameter is from icmp6, decode it. */
2502 ip6cp = (struct ip6ctlparam *)d;
2503 icmp6 = ip6cp->ip6c_icmp6;
2505 ip6 = ip6cp->ip6c_ip6;
2506 off = ip6cp->ip6c_off;
2507 sa6_src = ip6cp->ip6c_src;
2508 dst = ip6cp->ip6c_finaldst;
2512 off = 0; /* fool gcc */
2517 if (cmd == PRC_MSGSIZE)
2518 notify = tcp_mtudisc_notify;
2519 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2520 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2521 cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
2522 notify = tcp_drop_syn_sent;
2525 * Hostdead is ugly because it goes linearly through all PCBs.
2526 * XXX: We never get this from ICMP, otherwise it makes an
2527 * excellent DoS attack on machines with many connections.
2529 else if (cmd == PRC_HOSTDEAD)
2531 else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
2535 in6_pcbnotify(&V_tcbinfo, sa, 0,
2536 (const struct sockaddr *)sa6_src,
2537 0, cmd, NULL, notify);
2541 /* Check if we can safely get the ports from the tcp hdr */
2544 (int32_t) (off + sizeof(struct tcp_ports)))) {
2547 bzero(&t_ports, sizeof(struct tcp_ports));
2548 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
2549 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
2550 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
2551 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
2552 if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2553 /* signal EHOSTDOWN, as it flushes the cached route */
2554 inp = (*notify)(inp, EHOSTDOWN);
2557 off += sizeof(struct tcp_ports);
2558 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
2561 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
2563 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2564 !(inp->inp_flags & INP_DROPPED) &&
2565 !(inp->inp_socket == NULL)) {
2566 tp = intotcpcb(inp);
2567 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2568 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2569 if (cmd == PRC_MSGSIZE) {
2572 * If we got a needfrag set the MTU
2573 * in the route to the suggested new
2574 * value (if given) and then notify.
2576 mtu = ntohl(icmp6->icmp6_mtu);
2578 * If no alternative MTU was
2579 * proposed, or the proposed
2580 * MTU was too small, set to
2583 if (mtu < IPV6_MMTU)
2584 mtu = IPV6_MMTU - 8;
2585 bzero(&inc, sizeof(inc));
2586 inc.inc_fibnum = M_GETFIB(m);
2587 inc.inc_flags |= INC_ISIPV6;
2588 inc.inc6_faddr = *dst;
2589 if (in6_setscope(&inc.inc6_faddr,
2590 m->m_pkthdr.rcvif, NULL))
2593 * Only process the offered MTU if it
2594 * is smaller than the current one.
2596 if (mtu < tp->t_maxseg +
2597 sizeof (struct tcphdr) +
2598 sizeof (struct ip6_hdr)) {
2599 tcp_hc_updatemtu(&inc, mtu);
2600 tcp_mtudisc(inp, mtu);
2601 ICMP6STAT_INC(icp6s_pmtuchg);
2604 inp = (*notify)(inp,
2605 inet6ctlerrmap[cmd]);
2609 bzero(&inc, sizeof(inc));
2610 inc.inc_fibnum = M_GETFIB(m);
2611 inc.inc_flags |= INC_ISIPV6;
2612 inc.inc_fport = t_ports.th_dport;
2613 inc.inc_lport = t_ports.th_sport;
2614 inc.inc6_faddr = *dst;
2615 inc.inc6_laddr = ip6->ip6_src;
2616 syncache_unreach(&inc, icmp_tcp_seq);
2621 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2626 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len)
2632 MD5Update(&ctx, &inc->inc_fport, sizeof(uint16_t));
2633 MD5Update(&ctx, &inc->inc_lport, sizeof(uint16_t));
2634 switch (inc->inc_flags & INC_ISIPV6) {
2637 MD5Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr));
2638 MD5Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr));
2643 MD5Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr));
2644 MD5Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr));
2648 MD5Update(&ctx, key, len);
2649 MD5Final((unsigned char *)hash, &ctx);
2655 tcp_new_ts_offset(struct in_conninfo *inc)
2657 struct in_conninfo inc_store, *local_inc;
2659 if (!V_tcp_ts_offset_per_conn) {
2660 memcpy(&inc_store, inc, sizeof(struct in_conninfo));
2661 inc_store.inc_lport = 0;
2662 inc_store.inc_fport = 0;
2663 local_inc = &inc_store;
2667 return (tcp_keyed_hash(local_inc, V_ts_offset_secret,
2668 sizeof(V_ts_offset_secret)));
2672 * Following is where TCP initial sequence number generation occurs.
2674 * There are two places where we must use initial sequence numbers:
2675 * 1. In SYN-ACK packets.
2676 * 2. In SYN packets.
2678 * All ISNs for SYN-ACK packets are generated by the syncache. See
2679 * tcp_syncache.c for details.
2681 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2682 * depends on this property. In addition, these ISNs should be
2683 * unguessable so as to prevent connection hijacking. To satisfy
2684 * the requirements of this situation, the algorithm outlined in
2685 * RFC 1948 is used, with only small modifications.
2687 * Implementation details:
2689 * Time is based off the system timer, and is corrected so that it
2690 * increases by one megabyte per second. This allows for proper
2691 * recycling on high speed LANs while still leaving over an hour
2694 * As reading the *exact* system time is too expensive to be done
2695 * whenever setting up a TCP connection, we increment the time
2696 * offset in two ways. First, a small random positive increment
2697 * is added to isn_offset for each connection that is set up.
2698 * Second, the function tcp_isn_tick fires once per clock tick
2699 * and increments isn_offset as necessary so that sequence numbers
2700 * are incremented at approximately ISN_BYTES_PER_SECOND. The
2701 * random positive increments serve only to ensure that the same
2702 * exact sequence number is never sent out twice (as could otherwise
2703 * happen when a port is recycled in less than the system tick
2706 * net.inet.tcp.isn_reseed_interval controls the number of seconds
2707 * between seeding of isn_secret. This is normally set to zero,
2708 * as reseeding should not be necessary.
2710 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2711 * isn_offset_old, and isn_ctx is performed using the ISN lock. In
2712 * general, this means holding an exclusive (write) lock.
2715 #define ISN_BYTES_PER_SECOND 1048576
2716 #define ISN_STATIC_INCREMENT 4096
2717 #define ISN_RANDOM_INCREMENT (4096 - 1)
2718 #define ISN_SECRET_LENGTH 32
2720 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]);
2721 VNET_DEFINE_STATIC(int, isn_last);
2722 VNET_DEFINE_STATIC(int, isn_last_reseed);
2723 VNET_DEFINE_STATIC(u_int32_t, isn_offset);
2724 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old);
2726 #define V_isn_secret VNET(isn_secret)
2727 #define V_isn_last VNET(isn_last)
2728 #define V_isn_last_reseed VNET(isn_last_reseed)
2729 #define V_isn_offset VNET(isn_offset)
2730 #define V_isn_offset_old VNET(isn_offset_old)
2733 tcp_new_isn(struct in_conninfo *inc)
2736 u_int32_t projected_offset;
2739 /* Seed if this is the first use, reseed if requested. */
2740 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2741 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2743 arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0);
2744 V_isn_last_reseed = ticks;
2747 /* Compute the md5 hash and return the ISN. */
2748 new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret,
2749 sizeof(V_isn_secret));
2750 V_isn_offset += ISN_STATIC_INCREMENT +
2751 (arc4random() & ISN_RANDOM_INCREMENT);
2752 if (ticks != V_isn_last) {
2753 projected_offset = V_isn_offset_old +
2754 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2755 if (SEQ_GT(projected_offset, V_isn_offset))
2756 V_isn_offset = projected_offset;
2757 V_isn_offset_old = V_isn_offset;
2760 new_isn += V_isn_offset;
2766 * When a specific ICMP unreachable message is received and the
2767 * connection state is SYN-SENT, drop the connection. This behavior
2768 * is controlled by the icmp_may_rst sysctl.
2771 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2775 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2776 INP_WLOCK_ASSERT(inp);
2778 if ((inp->inp_flags & INP_TIMEWAIT) ||
2779 (inp->inp_flags & INP_DROPPED))
2782 tp = intotcpcb(inp);
2783 if (tp->t_state != TCPS_SYN_SENT)
2786 if (IS_FASTOPEN(tp->t_flags))
2787 tcp_fastopen_disable_path(tp);
2789 tp = tcp_drop(tp, errno);
2797 * When `need fragmentation' ICMP is received, update our idea of the MSS
2798 * based on the new value. Also nudge TCP to send something, since we
2799 * know the packet we just sent was dropped.
2800 * This duplicates some code in the tcp_mss() function in tcp_input.c.
2802 static struct inpcb *
2803 tcp_mtudisc_notify(struct inpcb *inp, int error)
2806 tcp_mtudisc(inp, -1);
2811 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2816 INP_WLOCK_ASSERT(inp);
2817 if ((inp->inp_flags & INP_TIMEWAIT) ||
2818 (inp->inp_flags & INP_DROPPED))
2821 tp = intotcpcb(inp);
2822 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2824 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2826 so = inp->inp_socket;
2827 SOCKBUF_LOCK(&so->so_snd);
2828 /* If the mss is larger than the socket buffer, decrease the mss. */
2829 if (so->so_snd.sb_hiwat < tp->t_maxseg)
2830 tp->t_maxseg = so->so_snd.sb_hiwat;
2831 SOCKBUF_UNLOCK(&so->so_snd);
2833 TCPSTAT_INC(tcps_mturesent);
2835 tp->snd_nxt = tp->snd_una;
2836 tcp_free_sackholes(tp);
2837 tp->snd_recover = tp->snd_max;
2838 if (tp->t_flags & TF_SACK_PERMIT)
2839 EXIT_FASTRECOVERY(tp->t_flags);
2840 tp->t_fb->tfb_tcp_output(tp);
2845 * Look-up the routing entry to the peer of this inpcb. If no route
2846 * is found and it cannot be allocated, then return 0. This routine
2847 * is called by TCP routines that access the rmx structure and by
2848 * tcp_mss_update to get the peer/interface MTU.
2851 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2853 struct nhop4_extended nh4;
2855 uint32_t maxmtu = 0;
2857 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2859 if (inc->inc_faddr.s_addr != INADDR_ANY) {
2861 if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr,
2862 NHR_REF, 0, &nh4) != 0)
2866 maxmtu = nh4.nh_mtu;
2868 /* Report additional interface capabilities. */
2870 if (ifp->if_capenable & IFCAP_TSO4 &&
2871 ifp->if_hwassist & CSUM_TSO) {
2872 cap->ifcap |= CSUM_TSO;
2873 cap->tsomax = ifp->if_hw_tsomax;
2874 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2875 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2878 fib4_free_nh_ext(inc->inc_fibnum, &nh4);
2886 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2888 struct nhop6_extended nh6;
2889 struct in6_addr dst6;
2892 uint32_t maxmtu = 0;
2894 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2896 if (inc->inc_flags & INC_IPV6MINMTU)
2899 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2900 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
2901 if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0,
2906 maxmtu = nh6.nh_mtu;
2908 /* Report additional interface capabilities. */
2910 if (ifp->if_capenable & IFCAP_TSO6 &&
2911 ifp->if_hwassist & CSUM_TSO) {
2912 cap->ifcap |= CSUM_TSO;
2913 cap->tsomax = ifp->if_hw_tsomax;
2914 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2915 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2918 fib6_free_nh_ext(inc->inc_fibnum, &nh6);
2926 * Calculate effective SMSS per RFC5681 definition for a given TCP
2927 * connection at its current state, taking into account SACK and etc.
2930 tcp_maxseg(const struct tcpcb *tp)
2934 if (tp->t_flags & TF_NOOPT)
2935 return (tp->t_maxseg);
2938 * Here we have a simplified code from tcp_addoptions(),
2939 * without a proper loop, and having most of paddings hardcoded.
2940 * We might make mistakes with padding here in some edge cases,
2941 * but this is harmless, since result of tcp_maxseg() is used
2942 * only in cwnd and ssthresh estimations.
2944 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4)
2945 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2946 if (tp->t_flags & TF_RCVD_TSTMP)
2947 optlen = TCPOLEN_TSTAMP_APPA;
2950 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2951 if (tp->t_flags & TF_SIGNATURE)
2952 optlen += PAD(TCPOLEN_SIGNATURE);
2954 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
2955 optlen += TCPOLEN_SACKHDR;
2956 optlen += tp->rcv_numsacks * TCPOLEN_SACK;
2957 optlen = PAD(optlen);
2960 if (tp->t_flags & TF_REQ_TSTMP)
2961 optlen = TCPOLEN_TSTAMP_APPA;
2963 optlen = PAD(TCPOLEN_MAXSEG);
2964 if (tp->t_flags & TF_REQ_SCALE)
2965 optlen += PAD(TCPOLEN_WINDOW);
2966 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2967 if (tp->t_flags & TF_SIGNATURE)
2968 optlen += PAD(TCPOLEN_SIGNATURE);
2970 if (tp->t_flags & TF_SACK_PERMIT)
2971 optlen += PAD(TCPOLEN_SACK_PERMITTED);
2974 optlen = min(optlen, TCP_MAXOLEN);
2975 return (tp->t_maxseg - optlen);
2979 sysctl_drop(SYSCTL_HANDLER_ARGS)
2981 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2982 struct sockaddr_storage addrs[2];
2986 struct sockaddr_in *fin, *lin;
2987 struct epoch_tracker et;
2989 struct sockaddr_in6 *fin6, *lin6;
3000 if (req->oldptr != NULL || req->oldlen != 0)
3002 if (req->newptr == NULL)
3004 if (req->newlen < sizeof(addrs))
3006 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3010 switch (addrs[0].ss_family) {
3013 fin6 = (struct sockaddr_in6 *)&addrs[0];
3014 lin6 = (struct sockaddr_in6 *)&addrs[1];
3015 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3016 lin6->sin6_len != sizeof(struct sockaddr_in6))
3018 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3019 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3021 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3022 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3023 fin = (struct sockaddr_in *)&addrs[0];
3024 lin = (struct sockaddr_in *)&addrs[1];
3027 error = sa6_embedscope(fin6, V_ip6_use_defzone);
3030 error = sa6_embedscope(lin6, V_ip6_use_defzone);
3037 fin = (struct sockaddr_in *)&addrs[0];
3038 lin = (struct sockaddr_in *)&addrs[1];
3039 if (fin->sin_len != sizeof(struct sockaddr_in) ||
3040 lin->sin_len != sizeof(struct sockaddr_in))
3047 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
3048 switch (addrs[0].ss_family) {
3051 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3052 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3053 INPLOOKUP_WLOCKPCB, NULL);
3058 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3059 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3064 if (inp->inp_flags & INP_TIMEWAIT) {
3066 * XXXRW: There currently exists a state where an
3067 * inpcb is present, but its timewait state has been
3068 * discarded. For now, don't allow dropping of this
3076 } else if (!(inp->inp_flags & INP_DROPPED) &&
3077 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
3078 tp = intotcpcb(inp);
3079 tp = tcp_drop(tp, ECONNABORTED);
3086 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
3090 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
3091 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
3092 0, sysctl_drop, "", "Drop TCP connection");
3095 * Generate a standardized TCP log line for use throughout the
3096 * tcp subsystem. Memory allocation is done with M_NOWAIT to
3097 * allow use in the interrupt context.
3099 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
3100 * NB: The function may return NULL if memory allocation failed.
3102 * Due to header inclusion and ordering limitations the struct ip
3103 * and ip6_hdr pointers have to be passed as void pointers.
3106 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3110 /* Is logging enabled? */
3111 if (tcp_log_in_vain == 0)
3114 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3118 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3122 /* Is logging enabled? */
3123 if (tcp_log_debug == 0)
3126 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3130 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3137 const struct ip6_hdr *ip6;
3139 ip6 = (const struct ip6_hdr *)ip6hdr;
3141 ip = (struct ip *)ip4hdr;
3144 * The log line looks like this:
3145 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
3147 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
3148 sizeof(PRINT_TH_FLAGS) + 1 +
3150 2 * INET6_ADDRSTRLEN;
3152 2 * INET_ADDRSTRLEN;
3155 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
3159 strcat(s, "TCP: [");
3162 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
3163 inet_ntoa_r(inc->inc_faddr, sp);
3165 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3167 inet_ntoa_r(inc->inc_laddr, sp);
3169 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3172 ip6_sprintf(sp, &inc->inc6_faddr);
3174 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3176 ip6_sprintf(sp, &inc->inc6_laddr);
3178 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3179 } else if (ip6 && th) {
3180 ip6_sprintf(sp, &ip6->ip6_src);
3182 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3184 ip6_sprintf(sp, &ip6->ip6_dst);
3186 sprintf(sp, "]:%i", ntohs(th->th_dport));
3189 } else if (ip && th) {
3190 inet_ntoa_r(ip->ip_src, sp);
3192 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3194 inet_ntoa_r(ip->ip_dst, sp);
3196 sprintf(sp, "]:%i", ntohs(th->th_dport));
3204 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
3205 if (*(s + size - 1) != '\0')
3206 panic("%s: string too long", __func__);
3211 * A subroutine which makes it easy to track TCP state changes with DTrace.
3212 * This function shouldn't be called for t_state initializations that don't
3213 * correspond to actual TCP state transitions.
3216 tcp_state_change(struct tcpcb *tp, int newstate)
3218 #if defined(KDTRACE_HOOKS)
3219 int pstate = tp->t_state;
3222 TCPSTATES_DEC(tp->t_state);
3223 TCPSTATES_INC(newstate);
3224 tp->t_state = newstate;
3225 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
3229 * Create an external-format (``xtcpcb'') structure using the information in
3230 * the kernel-format tcpcb structure pointed to by tp. This is done to
3231 * reduce the spew of irrelevant information over this interface, to isolate
3232 * user code from changes in the kernel structure, and potentially to provide
3233 * information-hiding if we decide that some of this information should be
3234 * hidden from users.
3237 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
3239 struct tcpcb *tp = intotcpcb(inp);
3242 bzero(xt, sizeof(*xt));
3243 if (inp->inp_flags & INP_TIMEWAIT) {
3244 xt->t_state = TCPS_TIME_WAIT;
3246 xt->t_state = tp->t_state;
3247 xt->t_logstate = tp->t_logstate;
3248 xt->t_flags = tp->t_flags;
3249 xt->t_sndzerowin = tp->t_sndzerowin;
3250 xt->t_sndrexmitpack = tp->t_sndrexmitpack;
3251 xt->t_rcvoopack = tp->t_rcvoopack;
3253 now = getsbinuptime();
3254 #define COPYTIMER(ttt) do { \
3255 if (callout_active(&tp->t_timers->ttt)) \
3256 xt->ttt = (tp->t_timers->ttt.c_time - now) / \
3261 COPYTIMER(tt_delack);
3262 COPYTIMER(tt_rexmt);
3263 COPYTIMER(tt_persist);
3267 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
3269 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
3270 TCP_FUNCTION_NAME_LEN_MAX);
3272 (void)tcp_log_get_id(tp, xt->xt_logid);
3276 xt->xt_len = sizeof(struct xtcpcb);
3277 in_pcbtoxinpcb(inp, &xt->xt_inp);
3278 if (inp->inp_socket == NULL)
3279 xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP;