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/cc/cc.h>
104 #include <netinet6/tcp6_var.h>
106 #include <netinet/tcpip.h>
107 #include <netinet/tcp_fastopen.h>
109 #include <netinet/tcp_pcap.h>
112 #include <netinet/tcp_debug.h>
115 #include <netinet6/ip6protosw.h>
118 #include <netinet/tcp_offload.h>
121 #include <netipsec/ipsec_support.h>
123 #include <machine/in_cksum.h>
126 #include <security/mac/mac_framework.h>
128 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
130 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
133 struct rwlock tcp_function_lock;
136 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
141 error = sysctl_handle_int(oidp, &new, 0, req);
142 if (error == 0 && req->newptr) {
143 if (new < TCP_MINMSS)
151 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
152 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
153 &sysctl_net_inet_tcp_mss_check, "I",
154 "Default TCP Maximum Segment Size");
158 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
162 new = V_tcp_v6mssdflt;
163 error = sysctl_handle_int(oidp, &new, 0, req);
164 if (error == 0 && req->newptr) {
165 if (new < TCP_MINMSS)
168 V_tcp_v6mssdflt = new;
173 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
174 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
175 &sysctl_net_inet_tcp_mss_v6_check, "I",
176 "Default TCP Maximum Segment Size for IPv6");
180 * Minimum MSS we accept and use. This prevents DoS attacks where
181 * we are forced to a ridiculous low MSS like 20 and send hundreds
182 * of packets instead of one. The effect scales with the available
183 * bandwidth and quickly saturates the CPU and network interface
184 * with packet generation and sending. Set to zero to disable MINMSS
185 * checking. This setting prevents us from sending too small packets.
187 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
188 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
189 &VNET_NAME(tcp_minmss), 0,
190 "Minimum TCP Maximum Segment Size");
192 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
193 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
194 &VNET_NAME(tcp_do_rfc1323), 0,
195 "Enable rfc1323 (high performance TCP) extensions");
197 static int tcp_log_debug = 0;
198 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
199 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
201 static int tcp_tcbhashsize;
202 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
203 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
205 static int do_tcpdrain = 1;
206 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
207 "Enable tcp_drain routine for extra help when low on mbufs");
209 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
210 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
212 static VNET_DEFINE(int, icmp_may_rst) = 1;
213 #define V_icmp_may_rst VNET(icmp_may_rst)
214 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
215 &VNET_NAME(icmp_may_rst), 0,
216 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
218 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
219 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
220 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
221 &VNET_NAME(tcp_isn_reseed_interval), 0,
222 "Seconds between reseeding of ISN secret");
224 static int tcp_soreceive_stream;
225 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
226 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
228 VNET_DEFINE(uma_zone_t, sack_hole_zone);
229 #define V_sack_hole_zone VNET(sack_hole_zone)
232 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
235 static struct inpcb *tcp_notify(struct inpcb *, int);
236 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
237 static void tcp_mtudisc(struct inpcb *, int);
238 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
239 void *ip4hdr, const void *ip6hdr);
242 static struct tcp_function_block tcp_def_funcblk = {
246 tcp_default_ctloutput,
257 int t_functions_inited = 0;
258 static int tcp_fb_cnt = 0;
259 struct tcp_funchead t_functions;
260 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
263 init_tcp_functions(void)
265 if (t_functions_inited == 0) {
266 TAILQ_INIT(&t_functions);
267 rw_init_flags(&tcp_function_lock, "tcp_func_lock" , 0);
268 t_functions_inited = 1;
272 static struct tcp_function_block *
273 find_tcp_functions_locked(struct tcp_function_set *fs)
275 struct tcp_function *f;
276 struct tcp_function_block *blk=NULL;
278 TAILQ_FOREACH(f, &t_functions, tf_next) {
279 if (strcmp(f->tf_name, fs->function_set_name) == 0) {
287 static struct tcp_function_block *
288 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
290 struct tcp_function_block *rblk=NULL;
291 struct tcp_function *f;
293 TAILQ_FOREACH(f, &t_functions, tf_next) {
294 if (f->tf_fb == blk) {
305 struct tcp_function_block *
306 find_and_ref_tcp_functions(struct tcp_function_set *fs)
308 struct tcp_function_block *blk;
310 rw_rlock(&tcp_function_lock);
311 blk = find_tcp_functions_locked(fs);
313 refcount_acquire(&blk->tfb_refcnt);
314 rw_runlock(&tcp_function_lock);
318 struct tcp_function_block *
319 find_and_ref_tcp_fb(struct tcp_function_block *blk)
321 struct tcp_function_block *rblk;
323 rw_rlock(&tcp_function_lock);
324 rblk = find_tcp_fb_locked(blk, NULL);
326 refcount_acquire(&rblk->tfb_refcnt);
327 rw_runlock(&tcp_function_lock);
333 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
336 struct tcp_function_set fs;
337 struct tcp_function_block *blk;
339 memset(&fs, 0, sizeof(fs));
340 rw_rlock(&tcp_function_lock);
341 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
344 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
345 fs.pcbcnt = blk->tfb_refcnt;
347 rw_runlock(&tcp_function_lock);
348 error = sysctl_handle_string(oidp, fs.function_set_name,
349 sizeof(fs.function_set_name), req);
351 /* Check for error or no change */
352 if (error != 0 || req->newptr == NULL)
355 rw_wlock(&tcp_function_lock);
356 blk = find_tcp_functions_locked(&fs);
358 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
362 tcp_func_set_ptr = blk;
364 rw_wunlock(&tcp_function_lock);
368 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
369 CTLTYPE_STRING | CTLFLAG_RW,
370 NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
371 "Set/get the default TCP functions");
374 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
376 int error, cnt, linesz;
377 struct tcp_function *f;
383 rw_rlock(&tcp_function_lock);
384 TAILQ_FOREACH(f, &t_functions, tf_next) {
387 rw_runlock(&tcp_function_lock);
389 bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
390 buffer = malloc(bufsz, M_TEMP, M_WAITOK);
395 linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
396 "Alias", "PCB count");
401 rw_rlock(&tcp_function_lock);
402 TAILQ_FOREACH(f, &t_functions, tf_next) {
403 alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
404 linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
405 f->tf_fb->tfb_tcp_block_name,
406 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
407 alias ? f->tf_name : "-",
408 f->tf_fb->tfb_refcnt);
409 if (linesz >= bufsz) {
417 rw_runlock(&tcp_function_lock);
419 error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
420 free(buffer, M_TEMP);
424 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
425 CTLTYPE_STRING|CTLFLAG_RD,
426 NULL, 0, sysctl_net_inet_list_available, "A",
427 "list available TCP Function sets");
430 * Exports one (struct tcp_function_info) for each alias/name.
433 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS)
436 struct tcp_function *f;
437 struct tcp_function_info tfi;
440 * We don't allow writes.
442 if (req->newptr != NULL)
446 * Wire the old buffer so we can directly copy the functions to
447 * user space without dropping the lock.
449 if (req->oldptr != NULL) {
450 error = sysctl_wire_old_buffer(req, 0);
456 * Walk the list and copy out matching entries. If INVARIANTS
457 * is compiled in, also walk the list to verify the length of
458 * the list matches what we have recorded.
460 rw_rlock(&tcp_function_lock);
464 if (req->oldptr == NULL) {
469 TAILQ_FOREACH(f, &t_functions, tf_next) {
473 if (req->oldptr != NULL) {
474 tfi.tfi_refcnt = f->tf_fb->tfb_refcnt;
475 tfi.tfi_id = f->tf_fb->tfb_id;
476 (void)strncpy(tfi.tfi_alias, f->tf_name,
477 TCP_FUNCTION_NAME_LEN_MAX);
478 tfi.tfi_alias[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
479 (void)strncpy(tfi.tfi_name,
480 f->tf_fb->tfb_tcp_block_name,
481 TCP_FUNCTION_NAME_LEN_MAX);
482 tfi.tfi_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
483 error = SYSCTL_OUT(req, &tfi, sizeof(tfi));
485 * Don't stop on error, as that is the
486 * mechanism we use to accumulate length
487 * information if the buffer was too short.
491 KASSERT(cnt == tcp_fb_cnt,
492 ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt));
496 rw_runlock(&tcp_function_lock);
497 if (req->oldptr == NULL)
498 error = SYSCTL_OUT(req, NULL,
499 (cnt + 1) * sizeof(struct tcp_function_info));
504 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info,
505 CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE,
506 NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info",
507 "List TCP function block name-to-ID mappings");
510 * Target size of TCP PCB hash tables. Must be a power of two.
512 * Note that this can be overridden by the kernel environment
513 * variable net.inet.tcp.tcbhashsize
516 #define TCBHASHSIZE 0
521 * Callouts should be moved into struct tcp directly. They are currently
522 * separate because the tcpcb structure is exported to userland for sysctl
523 * parsing purposes, which do not know about callouts.
534 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
535 #define V_tcpcb_zone VNET(tcpcb_zone)
537 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
538 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
540 static struct mtx isn_mtx;
542 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
543 #define ISN_LOCK() mtx_lock(&isn_mtx)
544 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
547 * TCP initialization.
550 tcp_zone_change(void *tag)
553 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
554 uma_zone_set_max(V_tcpcb_zone, maxsockets);
555 tcp_tw_zone_change();
559 tcp_inpcb_init(void *mem, int size, int flags)
561 struct inpcb *inp = mem;
563 INP_LOCK_INIT(inp, "inp", "tcpinp");
568 * Take a value and get the next power of 2 that doesn't overflow.
569 * Used to size the tcp_inpcb hash buckets.
572 maketcp_hashsize(int size)
578 * get the next power of 2 higher than maxsockets.
580 hashsize = 1 << fls(size);
581 /* catch overflow, and just go one power of 2 smaller */
582 if (hashsize < size) {
583 hashsize = 1 << (fls(size) - 1);
588 static volatile int next_tcp_stack_id = 1;
591 * Register a TCP function block with the name provided in the names
592 * array. (Note that this function does NOT automatically register
593 * blk->tfb_tcp_block_name as a stack name. Therefore, you should
594 * explicitly include blk->tfb_tcp_block_name in the list of names if
595 * you wish to register the stack with that name.)
597 * Either all name registrations will succeed or all will fail. If
598 * a name registration fails, the function will update the num_names
599 * argument to point to the array index of the name that encountered
602 * Returns 0 on success, or an error code on failure.
605 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
606 const char *names[], int *num_names)
608 struct tcp_function *n;
609 struct tcp_function_set fs;
612 KASSERT(names != NULL && *num_names > 0,
613 ("%s: Called with 0-length name list", __func__));
614 KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
616 if (t_functions_inited == 0) {
617 init_tcp_functions();
619 if ((blk->tfb_tcp_output == NULL) ||
620 (blk->tfb_tcp_do_segment == NULL) ||
621 (blk->tfb_tcp_ctloutput == NULL) ||
622 (strlen(blk->tfb_tcp_block_name) == 0)) {
624 * These functions are required and you
630 if (blk->tfb_tcp_timer_stop_all ||
631 blk->tfb_tcp_timer_activate ||
632 blk->tfb_tcp_timer_active ||
633 blk->tfb_tcp_timer_stop) {
635 * If you define one timer function you
636 * must have them all.
638 if ((blk->tfb_tcp_timer_stop_all == NULL) ||
639 (blk->tfb_tcp_timer_activate == NULL) ||
640 (blk->tfb_tcp_timer_active == NULL) ||
641 (blk->tfb_tcp_timer_stop == NULL)) {
647 refcount_init(&blk->tfb_refcnt, 0);
649 blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1);
650 for (i = 0; i < *num_names; i++) {
651 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
658 (void)strncpy(fs.function_set_name, names[i],
659 TCP_FUNCTION_NAME_LEN_MAX);
660 fs.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
661 rw_wlock(&tcp_function_lock);
662 if (find_tcp_functions_locked(&fs) != NULL) {
663 /* Duplicate name space not allowed */
664 rw_wunlock(&tcp_function_lock);
665 free(n, M_TCPFUNCTIONS);
669 (void)strncpy(n->tf_name, names[i], TCP_FUNCTION_NAME_LEN_MAX);
670 n->tf_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
671 TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
673 rw_wunlock(&tcp_function_lock);
679 * Deregister the names we just added. Because registration failed
680 * for names[i], we don't need to deregister that name.
683 rw_wlock(&tcp_function_lock);
685 TAILQ_FOREACH(n, &t_functions, tf_next) {
686 if (!strncmp(n->tf_name, names[i],
687 TCP_FUNCTION_NAME_LEN_MAX)) {
688 TAILQ_REMOVE(&t_functions, n, tf_next);
691 free(n, M_TCPFUNCTIONS);
696 rw_wunlock(&tcp_function_lock);
701 * Register a TCP function block using the name provided in the name
704 * Returns 0 on success, or an error code on failure.
707 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
710 const char *name_list[1];
717 name_list[0] = blk->tfb_tcp_block_name;
718 rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
723 * Register a TCP function block using the name defined in
724 * blk->tfb_tcp_block_name.
726 * Returns 0 on success, or an error code on failure.
729 register_tcp_functions(struct tcp_function_block *blk, int wait)
732 return (register_tcp_functions_as_name(blk, NULL, wait));
736 deregister_tcp_functions(struct tcp_function_block *blk)
738 struct tcp_function *f;
741 if (strcmp(blk->tfb_tcp_block_name, "default") == 0) {
742 /* You can't un-register the default */
745 rw_wlock(&tcp_function_lock);
746 if (blk == tcp_func_set_ptr) {
747 /* You can't free the current default */
748 rw_wunlock(&tcp_function_lock);
751 if (blk->tfb_refcnt) {
752 /* Still tcb attached, mark it. */
753 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
754 rw_wunlock(&tcp_function_lock);
757 while (find_tcp_fb_locked(blk, &f) != NULL) {
759 TAILQ_REMOVE(&t_functions, f, tf_next);
762 free(f, M_TCPFUNCTIONS);
765 rw_wunlock(&tcp_function_lock);
772 const char *tcbhash_tuneable;
775 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
778 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
779 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
780 printf("%s: WARNING: unable to register helper hook\n", __func__);
781 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
782 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
783 printf("%s: WARNING: unable to register helper hook\n", __func__);
785 hashsize = TCBHASHSIZE;
786 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
789 * Auto tune the hash size based on maxsockets.
790 * A perfect hash would have a 1:1 mapping
791 * (hashsize = maxsockets) however it's been
792 * suggested that O(2) average is better.
794 hashsize = maketcp_hashsize(maxsockets / 4);
796 * Our historical default is 512,
797 * do not autotune lower than this.
801 if (bootverbose && IS_DEFAULT_VNET(curvnet))
802 printf("%s: %s auto tuned to %d\n", __func__,
803 tcbhash_tuneable, hashsize);
806 * We require a hashsize to be a power of two.
807 * Previously if it was not a power of two we would just reset it
808 * back to 512, which could be a nasty surprise if you did not notice
810 * Instead what we do is clip it to the closest power of two lower
811 * than the specified hash value.
813 if (!powerof2(hashsize)) {
814 int oldhashsize = hashsize;
816 hashsize = maketcp_hashsize(hashsize);
817 /* prevent absurdly low value */
820 printf("%s: WARNING: TCB hash size not a power of 2, "
821 "clipped from %d to %d.\n", __func__, oldhashsize,
824 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
825 "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE);
828 * These have to be type stable for the benefit of the timers.
830 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
831 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
832 uma_zone_set_max(V_tcpcb_zone, maxsockets);
833 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
839 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
840 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
841 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
845 /* Skip initialization of globals for non-default instances. */
846 if (!IS_DEFAULT_VNET(curvnet))
849 tcp_reass_global_init();
851 /* XXX virtualize those bellow? */
852 tcp_delacktime = TCPTV_DELACK;
853 tcp_keepinit = TCPTV_KEEP_INIT;
854 tcp_keepidle = TCPTV_KEEP_IDLE;
855 tcp_keepintvl = TCPTV_KEEPINTVL;
856 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
858 tcp_rexmit_min = TCPTV_MIN;
859 if (tcp_rexmit_min < 1)
861 tcp_persmin = TCPTV_PERSMIN;
862 tcp_persmax = TCPTV_PERSMAX;
863 tcp_rexmit_slop = TCPTV_CPU_VAR;
864 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
865 tcp_tcbhashsize = hashsize;
866 /* Setup the tcp function block list */
867 init_tcp_functions();
868 register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
870 /* Initialize the TCP logging data. */
874 if (tcp_soreceive_stream) {
876 tcp_usrreqs.pru_soreceive = soreceive_stream;
879 tcp6_usrreqs.pru_soreceive = soreceive_stream;
884 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
886 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
888 if (max_protohdr < TCP_MINPROTOHDR)
889 max_protohdr = TCP_MINPROTOHDR;
890 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
892 #undef TCP_MINPROTOHDR
895 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
896 SHUTDOWN_PRI_DEFAULT);
897 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
898 EVENTHANDLER_PRI_ANY);
906 tcp_destroy(void *unused __unused)
914 * All our processes are gone, all our sockets should be cleaned
915 * up, which means, we should be past the tcp_discardcb() calls.
916 * Sleep to let all tcpcb timers really disappear and cleanup.
919 INP_LIST_RLOCK(&V_tcbinfo);
920 n = V_tcbinfo.ipi_count;
921 INP_LIST_RUNLOCK(&V_tcbinfo);
924 pause("tcpdes", hz / 10);
929 in_pcbinfo_destroy(&V_tcbinfo);
930 /* tcp_discardcb() clears the sack_holes up. */
931 uma_zdestroy(V_sack_hole_zone);
932 uma_zdestroy(V_tcpcb_zone);
935 * Cannot free the zone until all tcpcbs are released as we attach
936 * the allocations to them.
938 tcp_fastopen_destroy();
941 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
943 printf("%s: WARNING: unable to deregister helper hook "
944 "type=%d, id=%d: error %d returned\n", __func__,
945 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
947 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
949 printf("%s: WARNING: unable to deregister helper hook "
950 "type=%d, id=%d: error %d returned\n", __func__,
951 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
955 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
965 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
966 * tcp_template used to store this data in mbufs, but we now recopy it out
967 * of the tcpcb each time to conserve mbufs.
970 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
972 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
974 INP_WLOCK_ASSERT(inp);
977 if ((inp->inp_vflag & INP_IPV6) != 0) {
980 ip6 = (struct ip6_hdr *)ip_ptr;
981 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
982 (inp->inp_flow & IPV6_FLOWINFO_MASK);
983 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
984 (IPV6_VERSION & IPV6_VERSION_MASK);
985 ip6->ip6_nxt = IPPROTO_TCP;
986 ip6->ip6_plen = htons(sizeof(struct tcphdr));
987 ip6->ip6_src = inp->in6p_laddr;
988 ip6->ip6_dst = inp->in6p_faddr;
991 #if defined(INET6) && defined(INET)
998 ip = (struct ip *)ip_ptr;
999 ip->ip_v = IPVERSION;
1001 ip->ip_tos = inp->inp_ip_tos;
1005 ip->ip_ttl = inp->inp_ip_ttl;
1007 ip->ip_p = IPPROTO_TCP;
1008 ip->ip_src = inp->inp_laddr;
1009 ip->ip_dst = inp->inp_faddr;
1012 th->th_sport = inp->inp_lport;
1013 th->th_dport = inp->inp_fport;
1021 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
1025 * Create template to be used to send tcp packets on a connection.
1026 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
1027 * use for this function is in keepalives, which use tcp_respond.
1030 tcpip_maketemplate(struct inpcb *inp)
1034 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
1037 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
1042 * Send a single message to the TCP at address specified by
1043 * the given TCP/IP header. If m == NULL, then we make a copy
1044 * of the tcpiphdr at th and send directly to the addressed host.
1045 * This is used to force keep alive messages out using the TCP
1046 * template for a connection. If flags are given then we send
1047 * a message back to the TCP which originated the segment th,
1048 * and discard the mbuf containing it and any other attached mbufs.
1050 * In any case the ack and sequence number of the transmitted
1051 * segment are as specified by the parameters.
1053 * NOTE: If m != NULL, then th must point to *inside* the mbuf.
1056 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
1057 tcp_seq ack, tcp_seq seq, int flags)
1066 struct ip6_hdr *ip6;
1069 int optlen, tlen, win;
1072 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
1075 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
1082 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
1083 INP_WLOCK_ASSERT(inp);
1090 if (!(flags & TH_RST)) {
1091 win = sbspace(&inp->inp_socket->so_rcv);
1092 if (win > TCP_MAXWIN << tp->rcv_scale)
1093 win = TCP_MAXWIN << tp->rcv_scale;
1095 if ((tp->t_flags & TF_NOOPT) == 0)
1099 m = m_gethdr(M_NOWAIT, MT_DATA);
1102 m->m_data += max_linkhdr;
1105 bcopy((caddr_t)ip6, mtod(m, caddr_t),
1106 sizeof(struct ip6_hdr));
1107 ip6 = mtod(m, struct ip6_hdr *);
1108 nth = (struct tcphdr *)(ip6 + 1);
1112 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1113 ip = mtod(m, struct ip *);
1114 nth = (struct tcphdr *)(ip + 1);
1116 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1118 } else if (!M_WRITABLE(m)) {
1121 /* Can't reuse 'm', allocate a new mbuf. */
1122 n = m_gethdr(M_NOWAIT, MT_DATA);
1128 if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1134 n->m_data += max_linkhdr;
1135 /* m_len is set later */
1136 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1139 bcopy((caddr_t)ip6, mtod(n, caddr_t),
1140 sizeof(struct ip6_hdr));
1141 ip6 = mtod(n, struct ip6_hdr *);
1142 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1143 nth = (struct tcphdr *)(ip6 + 1);
1147 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1148 ip = mtod(n, struct ip *);
1149 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1150 nth = (struct tcphdr *)(ip + 1);
1152 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1153 xchg(nth->th_dport, nth->th_sport, uint16_t);
1160 * XXX MRT We inherit the FIB, which is lucky.
1164 m->m_data = (caddr_t)ipgen;
1165 /* m_len is set later */
1168 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1169 nth = (struct tcphdr *)(ip6 + 1);
1173 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1174 nth = (struct tcphdr *)(ip + 1);
1178 * this is usually a case when an extension header
1179 * exists between the IPv6 header and the
1182 nth->th_sport = th->th_sport;
1183 nth->th_dport = th->th_dport;
1185 xchg(nth->th_dport, nth->th_sport, uint16_t);
1191 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1193 #if defined(INET) && defined(INET6)
1197 tlen = sizeof (struct tcpiphdr);
1201 KASSERT(M_TRAILINGSPACE(m) >= tlen,
1202 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1203 m, tlen, (long)M_TRAILINGSPACE(m)));
1208 /* Make sure we have room. */
1209 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1210 m->m_next = m_get(M_NOWAIT, MT_DATA);
1212 optp = mtod(m->m_next, u_char *);
1217 optp = (u_char *) (nth + 1);
1223 if (tp->t_flags & TF_RCVD_TSTMP) {
1224 to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1225 to.to_tsecr = tp->ts_recent;
1226 to.to_flags |= TOF_TS;
1228 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1229 /* TCP-MD5 (RFC2385). */
1230 if (tp->t_flags & TF_SIGNATURE)
1231 to.to_flags |= TOF_SIGNATURE;
1233 /* Add the options. */
1234 tlen += optlen = tcp_addoptions(&to, optp);
1236 /* Update m_len in the correct mbuf. */
1237 optm->m_len += optlen;
1243 ip6->ip6_vfc = IPV6_VERSION;
1244 ip6->ip6_nxt = IPPROTO_TCP;
1245 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1248 #if defined(INET) && defined(INET6)
1253 ip->ip_len = htons(tlen);
1254 ip->ip_ttl = V_ip_defttl;
1255 if (V_path_mtu_discovery)
1256 ip->ip_off |= htons(IP_DF);
1259 m->m_pkthdr.len = tlen;
1260 m->m_pkthdr.rcvif = NULL;
1264 * Packet is associated with a socket, so allow the
1265 * label of the response to reflect the socket label.
1267 INP_WLOCK_ASSERT(inp);
1268 mac_inpcb_create_mbuf(inp, m);
1271 * Packet is not associated with a socket, so possibly
1272 * update the label in place.
1274 mac_netinet_tcp_reply(m);
1277 nth->th_seq = htonl(seq);
1278 nth->th_ack = htonl(ack);
1280 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1281 nth->th_flags = flags;
1283 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1285 nth->th_win = htons((u_short)win);
1288 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1289 if (to.to_flags & TOF_SIGNATURE) {
1290 if (!TCPMD5_ENABLED() ||
1291 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
1298 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1301 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1302 nth->th_sum = in6_cksum_pseudo(ip6,
1303 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
1304 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
1308 #if defined(INET6) && defined(INET)
1313 m->m_pkthdr.csum_flags = CSUM_TCP;
1314 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1315 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1319 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1320 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1322 TCP_PROBE3(debug__output, tp, th, m);
1324 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
1328 TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
1329 (void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
1332 #if defined(INET) && defined(INET6)
1337 TCP_PROBE5(send, NULL, tp, ip, tp, nth);
1338 (void)ip_output(m, NULL, NULL, 0, NULL, inp);
1344 * Create a new TCP control block, making an
1345 * empty reassembly queue and hooking it to the argument
1346 * protocol control block. The `inp' parameter must have
1347 * come from the zone allocator set up in tcp_init().
1350 tcp_newtcpcb(struct inpcb *inp)
1352 struct tcpcb_mem *tm;
1355 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1358 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1363 /* Initialise cc_var struct for this tcpcb. */
1365 tp->ccv->type = IPPROTO_TCP;
1366 tp->ccv->ccvc.tcp = tp;
1367 rw_rlock(&tcp_function_lock);
1368 tp->t_fb = tcp_func_set_ptr;
1369 refcount_acquire(&tp->t_fb->tfb_refcnt);
1370 rw_runlock(&tcp_function_lock);
1372 * Use the current system default CC algorithm.
1375 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1376 CC_ALGO(tp) = CC_DEFAULT();
1379 if (CC_ALGO(tp)->cb_init != NULL)
1380 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1381 if (tp->t_fb->tfb_tcp_fb_fini)
1382 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1383 refcount_release(&tp->t_fb->tfb_refcnt);
1384 uma_zfree(V_tcpcb_zone, tm);
1390 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1391 if (tp->t_fb->tfb_tcp_fb_fini)
1392 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1393 refcount_release(&tp->t_fb->tfb_refcnt);
1394 uma_zfree(V_tcpcb_zone, tm);
1400 tp->t_vnet = inp->inp_vnet;
1402 tp->t_timers = &tm->tt;
1403 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
1406 isipv6 ? V_tcp_v6mssdflt :
1410 /* Set up our timeouts. */
1411 callout_init(&tp->t_timers->tt_rexmt, 1);
1412 callout_init(&tp->t_timers->tt_persist, 1);
1413 callout_init(&tp->t_timers->tt_keep, 1);
1414 callout_init(&tp->t_timers->tt_2msl, 1);
1415 callout_init(&tp->t_timers->tt_delack, 1);
1417 if (V_tcp_do_rfc1323)
1418 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1420 tp->t_flags |= TF_SACK_PERMIT;
1421 TAILQ_INIT(&tp->snd_holes);
1423 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1426 in_pcbref(inp); /* Reference for tcpcb */
1430 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1431 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
1432 * reasonable initial retransmit time.
1434 tp->t_srtt = TCPTV_SRTTBASE;
1435 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1436 tp->t_rttmin = tcp_rexmit_min;
1437 tp->t_rxtcur = TCPTV_RTOBASE;
1438 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1439 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1440 tp->t_rcvtime = ticks;
1442 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1443 * because the socket may be bound to an IPv6 wildcard address,
1444 * which may match an IPv4-mapped IPv6 address.
1446 inp->inp_ip_ttl = V_ip_defttl;
1450 * Init the TCP PCAP queues.
1452 tcp_pcap_tcpcb_init(tp);
1455 /* Initialize the per-TCPCB log data. */
1456 tcp_log_tcpcbinit(tp);
1458 if (tp->t_fb->tfb_tcp_fb_init) {
1459 (*tp->t_fb->tfb_tcp_fb_init)(tp);
1461 return (tp); /* XXX */
1465 * Switch the congestion control algorithm back to NewReno for any active
1466 * control blocks using an algorithm which is about to go away.
1467 * This ensures the CC framework can allow the unload to proceed without leaving
1468 * any dangling pointers which would trigger a panic.
1469 * Returning non-zero would inform the CC framework that something went wrong
1470 * and it would be unsafe to allow the unload to proceed. However, there is no
1471 * way for this to occur with this implementation so we always return zero.
1474 tcp_ccalgounload(struct cc_algo *unload_algo)
1476 struct cc_algo *tmpalgo;
1479 VNET_ITERATOR_DECL(vnet_iter);
1482 * Check all active control blocks across all network stacks and change
1483 * any that are using "unload_algo" back to NewReno. If "unload_algo"
1484 * requires cleanup code to be run, call it.
1487 VNET_FOREACH(vnet_iter) {
1488 CURVNET_SET(vnet_iter);
1489 INP_INFO_WLOCK(&V_tcbinfo);
1491 * New connections already part way through being initialised
1492 * with the CC algo we're removing will not race with this code
1493 * because the INP_INFO_WLOCK is held during initialisation. We
1494 * therefore don't enter the loop below until the connection
1495 * list has stabilised.
1497 LIST_FOREACH(inp, &V_tcb, inp_list) {
1499 /* Important to skip tcptw structs. */
1500 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1501 (tp = intotcpcb(inp)) != NULL) {
1503 * By holding INP_WLOCK here, we are assured
1504 * that the connection is not currently
1505 * executing inside the CC module's functions
1506 * i.e. it is safe to make the switch back to
1509 if (CC_ALGO(tp) == unload_algo) {
1510 tmpalgo = CC_ALGO(tp);
1511 /* NewReno does not require any init. */
1512 CC_ALGO(tp) = &newreno_cc_algo;
1513 if (tmpalgo->cb_destroy != NULL)
1514 tmpalgo->cb_destroy(tp->ccv);
1519 INP_INFO_WUNLOCK(&V_tcbinfo);
1522 VNET_LIST_RUNLOCK();
1528 * Drop a TCP connection, reporting
1529 * the specified error. If connection is synchronized,
1530 * then send a RST to peer.
1533 tcp_drop(struct tcpcb *tp, int errno)
1535 struct socket *so = tp->t_inpcb->inp_socket;
1537 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1538 INP_WLOCK_ASSERT(tp->t_inpcb);
1540 if (TCPS_HAVERCVDSYN(tp->t_state)) {
1541 tcp_state_change(tp, TCPS_CLOSED);
1542 (void) tp->t_fb->tfb_tcp_output(tp);
1543 TCPSTAT_INC(tcps_drops);
1545 TCPSTAT_INC(tcps_conndrops);
1546 if (errno == ETIMEDOUT && tp->t_softerror)
1547 errno = tp->t_softerror;
1548 so->so_error = errno;
1549 return (tcp_close(tp));
1553 tcp_discardcb(struct tcpcb *tp)
1555 struct inpcb *inp = tp->t_inpcb;
1556 struct socket *so = inp->inp_socket;
1558 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1562 INP_WLOCK_ASSERT(inp);
1565 * Make sure that all of our timers are stopped before we delete the
1568 * If stopping a timer fails, we schedule a discard function in same
1569 * callout, and the last discard function called will take care of
1570 * deleting the tcpcb.
1572 tp->t_timers->tt_draincnt = 0;
1573 tcp_timer_stop(tp, TT_REXMT);
1574 tcp_timer_stop(tp, TT_PERSIST);
1575 tcp_timer_stop(tp, TT_KEEP);
1576 tcp_timer_stop(tp, TT_2MSL);
1577 tcp_timer_stop(tp, TT_DELACK);
1578 if (tp->t_fb->tfb_tcp_timer_stop_all) {
1580 * Call the stop-all function of the methods,
1581 * this function should call the tcp_timer_stop()
1582 * method with each of the function specific timeouts.
1583 * That stop will be called via the tfb_tcp_timer_stop()
1584 * which should use the async drain function of the
1585 * callout system (see tcp_var.h).
1587 tp->t_fb->tfb_tcp_timer_stop_all(tp);
1591 * If we got enough samples through the srtt filter,
1592 * save the rtt and rttvar in the routing entry.
1593 * 'Enough' is arbitrarily defined as 4 rtt samples.
1594 * 4 samples is enough for the srtt filter to converge
1595 * to within enough % of the correct value; fewer samples
1596 * and we could save a bogus rtt. The danger is not high
1597 * as tcp quickly recovers from everything.
1598 * XXX: Works very well but needs some more statistics!
1600 if (tp->t_rttupdated >= 4) {
1601 struct hc_metrics_lite metrics;
1604 bzero(&metrics, sizeof(metrics));
1606 * Update the ssthresh always when the conditions below
1607 * are satisfied. This gives us better new start value
1608 * for the congestion avoidance for new connections.
1609 * ssthresh is only set if packet loss occurred on a session.
1611 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1612 * being torn down. Ideally this code would not use 'so'.
1614 ssthresh = tp->snd_ssthresh;
1615 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1617 * convert the limit from user data bytes to
1618 * packets then to packet data bytes.
1620 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1623 ssthresh *= (tp->t_maxseg +
1625 (isipv6 ? sizeof (struct ip6_hdr) +
1626 sizeof (struct tcphdr) :
1628 sizeof (struct tcpiphdr)
1635 metrics.rmx_ssthresh = ssthresh;
1637 metrics.rmx_rtt = tp->t_srtt;
1638 metrics.rmx_rttvar = tp->t_rttvar;
1639 metrics.rmx_cwnd = tp->snd_cwnd;
1640 metrics.rmx_sendpipe = 0;
1641 metrics.rmx_recvpipe = 0;
1643 tcp_hc_update(&inp->inp_inc, &metrics);
1646 /* free the reassembly queue, if any */
1647 tcp_reass_flush(tp);
1650 /* Disconnect offload device, if any. */
1651 if (tp->t_flags & TF_TOE)
1652 tcp_offload_detach(tp);
1655 tcp_free_sackholes(tp);
1658 /* Free the TCP PCAP queues. */
1659 tcp_pcap_drain(&(tp->t_inpkts));
1660 tcp_pcap_drain(&(tp->t_outpkts));
1663 /* Allow the CC algorithm to clean up after itself. */
1664 if (CC_ALGO(tp)->cb_destroy != NULL)
1665 CC_ALGO(tp)->cb_destroy(tp->ccv);
1668 khelp_destroy_osd(tp->osd);
1672 inp->inp_ppcb = NULL;
1673 if (tp->t_timers->tt_draincnt == 0) {
1674 /* We own the last reference on tcpcb, let's free it. */
1676 tcp_log_tcpcbfini(tp);
1678 TCPSTATES_DEC(tp->t_state);
1679 if (tp->t_fb->tfb_tcp_fb_fini)
1680 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1681 refcount_release(&tp->t_fb->tfb_refcnt);
1683 uma_zfree(V_tcpcb_zone, tp);
1684 released = in_pcbrele_wlocked(inp);
1685 KASSERT(!released, ("%s: inp %p should not have been released "
1686 "here", __func__, inp));
1691 tcp_timer_discard(void *ptp)
1696 tp = (struct tcpcb *)ptp;
1697 CURVNET_SET(tp->t_vnet);
1698 INP_INFO_RLOCK(&V_tcbinfo);
1700 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1703 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1704 ("%s: tcpcb has to be stopped here", __func__));
1705 tp->t_timers->tt_draincnt--;
1706 if (tp->t_timers->tt_draincnt == 0) {
1707 /* We own the last reference on this tcpcb, let's free it. */
1709 tcp_log_tcpcbfini(tp);
1711 TCPSTATES_DEC(tp->t_state);
1712 if (tp->t_fb->tfb_tcp_fb_fini)
1713 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1714 refcount_release(&tp->t_fb->tfb_refcnt);
1716 uma_zfree(V_tcpcb_zone, tp);
1717 if (in_pcbrele_wlocked(inp)) {
1718 INP_INFO_RUNLOCK(&V_tcbinfo);
1724 INP_INFO_RUNLOCK(&V_tcbinfo);
1729 * Attempt to close a TCP control block, marking it as dropped, and freeing
1730 * the socket if we hold the only reference.
1733 tcp_close(struct tcpcb *tp)
1735 struct inpcb *inp = tp->t_inpcb;
1738 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1739 INP_WLOCK_ASSERT(inp);
1742 if (tp->t_state == TCPS_LISTEN)
1743 tcp_offload_listen_stop(tp);
1746 * This releases the TFO pending counter resource for TFO listen
1747 * sockets as well as passively-created TFO sockets that transition
1748 * from SYN_RECEIVED to CLOSED.
1750 if (tp->t_tfo_pending) {
1751 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
1752 tp->t_tfo_pending = NULL;
1755 TCPSTAT_INC(tcps_closed);
1756 if (tp->t_state != TCPS_CLOSED)
1757 tcp_state_change(tp, TCPS_CLOSED);
1758 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1759 so = inp->inp_socket;
1760 soisdisconnected(so);
1761 if (inp->inp_flags & INP_SOCKREF) {
1762 KASSERT(so->so_state & SS_PROTOREF,
1763 ("tcp_close: !SS_PROTOREF"));
1764 inp->inp_flags &= ~INP_SOCKREF;
1767 so->so_state &= ~SS_PROTOREF;
1777 VNET_ITERATOR_DECL(vnet_iter);
1782 VNET_LIST_RLOCK_NOSLEEP();
1783 VNET_FOREACH(vnet_iter) {
1784 CURVNET_SET(vnet_iter);
1789 * Walk the tcpbs, if existing, and flush the reassembly queue,
1790 * if there is one...
1791 * XXX: The "Net/3" implementation doesn't imply that the TCP
1792 * reassembly queue should be flushed, but in a situation
1793 * where we're really low on mbufs, this is potentially
1796 INP_INFO_WLOCK(&V_tcbinfo);
1797 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1798 if (inpb->inp_flags & INP_TIMEWAIT)
1801 if ((tcpb = intotcpcb(inpb)) != NULL) {
1802 tcp_reass_flush(tcpb);
1803 tcp_clean_sackreport(tcpb);
1805 tcp_log_drain(tcpb);
1808 if (tcp_pcap_aggressive_free) {
1809 /* Free the TCP PCAP queues. */
1810 tcp_pcap_drain(&(tcpb->t_inpkts));
1811 tcp_pcap_drain(&(tcpb->t_outpkts));
1817 INP_INFO_WUNLOCK(&V_tcbinfo);
1820 VNET_LIST_RUNLOCK_NOSLEEP();
1824 * Notify a tcp user of an asynchronous error;
1825 * store error as soft error, but wake up user
1826 * (for now, won't do anything until can select for soft error).
1828 * Do not wake up user since there currently is no mechanism for
1829 * reporting soft errors (yet - a kqueue filter may be added).
1831 static struct inpcb *
1832 tcp_notify(struct inpcb *inp, int error)
1836 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1837 INP_WLOCK_ASSERT(inp);
1839 if ((inp->inp_flags & INP_TIMEWAIT) ||
1840 (inp->inp_flags & INP_DROPPED))
1843 tp = intotcpcb(inp);
1844 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1847 * Ignore some errors if we are hooked up.
1848 * If connection hasn't completed, has retransmitted several times,
1849 * and receives a second error, give up now. This is better
1850 * than waiting a long time to establish a connection that
1851 * can never complete.
1853 if (tp->t_state == TCPS_ESTABLISHED &&
1854 (error == EHOSTUNREACH || error == ENETUNREACH ||
1855 error == EHOSTDOWN)) {
1856 if (inp->inp_route.ro_rt) {
1857 RTFREE(inp->inp_route.ro_rt);
1858 inp->inp_route.ro_rt = (struct rtentry *)NULL;
1861 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1863 tp = tcp_drop(tp, error);
1869 tp->t_softerror = error;
1873 wakeup( &so->so_timeo);
1880 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1882 int error, i, m, n, pcb_count;
1883 struct inpcb *inp, **inp_list;
1888 * The process of preparing the TCB list is too time-consuming and
1889 * resource-intensive to repeat twice on every request.
1891 if (req->oldptr == NULL) {
1892 n = V_tcbinfo.ipi_count +
1893 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
1894 n += imax(n / 8, 10);
1895 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1899 if (req->newptr != NULL)
1903 * OK, now we're committed to doing something.
1905 INP_LIST_RLOCK(&V_tcbinfo);
1906 gencnt = V_tcbinfo.ipi_gencnt;
1907 n = V_tcbinfo.ipi_count;
1908 INP_LIST_RUNLOCK(&V_tcbinfo);
1910 m = counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
1912 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1913 + (n + m) * sizeof(struct xtcpcb));
1917 xig.xig_len = sizeof xig;
1918 xig.xig_count = n + m;
1919 xig.xig_gen = gencnt;
1920 xig.xig_sogen = so_gencnt;
1921 error = SYSCTL_OUT(req, &xig, sizeof xig);
1925 error = syncache_pcblist(req, m, &pcb_count);
1929 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1931 INP_INFO_WLOCK(&V_tcbinfo);
1932 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1933 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1935 if (inp->inp_gencnt <= gencnt) {
1937 * XXX: This use of cr_cansee(), introduced with
1938 * TCP state changes, is not quite right, but for
1939 * now, better than nothing.
1941 if (inp->inp_flags & INP_TIMEWAIT) {
1942 if (intotw(inp) != NULL)
1943 error = cr_cansee(req->td->td_ucred,
1944 intotw(inp)->tw_cred);
1946 error = EINVAL; /* Skip this inp. */
1948 error = cr_canseeinpcb(req->td->td_ucred, inp);
1951 inp_list[i++] = inp;
1956 INP_INFO_WUNLOCK(&V_tcbinfo);
1960 for (i = 0; i < n; i++) {
1963 if (inp->inp_gencnt <= gencnt) {
1966 tcp_inptoxtp(inp, &xt);
1968 error = SYSCTL_OUT(req, &xt, sizeof xt);
1972 INP_INFO_RLOCK(&V_tcbinfo);
1973 for (i = 0; i < n; i++) {
1976 if (!in_pcbrele_rlocked(inp))
1979 INP_INFO_RUNLOCK(&V_tcbinfo);
1983 * Give the user an updated idea of our state.
1984 * If the generation differs from what we told
1985 * her before, she knows that something happened
1986 * while we were processing this request, and it
1987 * might be necessary to retry.
1989 INP_LIST_RLOCK(&V_tcbinfo);
1990 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1991 xig.xig_sogen = so_gencnt;
1992 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1993 INP_LIST_RUNLOCK(&V_tcbinfo);
1994 error = SYSCTL_OUT(req, &xig, sizeof xig);
1996 free(inp_list, M_TEMP);
2000 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
2001 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
2002 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
2006 tcp_getcred(SYSCTL_HANDLER_ARGS)
2009 struct sockaddr_in addrs[2];
2013 error = priv_check(req->td, PRIV_NETINET_GETCRED);
2016 error = SYSCTL_IN(req, addrs, sizeof(addrs));
2019 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
2020 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
2022 if (inp->inp_socket == NULL)
2025 error = cr_canseeinpcb(req->td->td_ucred, inp);
2027 cru2x(inp->inp_cred, &xuc);
2032 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2036 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
2037 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
2038 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
2043 tcp6_getcred(SYSCTL_HANDLER_ARGS)
2046 struct sockaddr_in6 addrs[2];
2053 error = priv_check(req->td, PRIV_NETINET_GETCRED);
2056 error = SYSCTL_IN(req, addrs, sizeof(addrs));
2059 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
2060 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
2063 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
2065 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
2074 inp = in_pcblookup(&V_tcbinfo,
2075 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
2077 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
2078 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
2081 inp = in6_pcblookup(&V_tcbinfo,
2082 &addrs[1].sin6_addr, addrs[1].sin6_port,
2083 &addrs[0].sin6_addr, addrs[0].sin6_port,
2084 INPLOOKUP_RLOCKPCB, NULL);
2086 if (inp->inp_socket == NULL)
2089 error = cr_canseeinpcb(req->td->td_ucred, inp);
2091 cru2x(inp->inp_cred, &xuc);
2096 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2100 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
2101 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
2102 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
2108 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
2110 struct ip *ip = vip;
2112 struct in_addr faddr;
2115 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2117 struct in_conninfo inc;
2118 tcp_seq icmp_tcp_seq;
2121 faddr = ((struct sockaddr_in *)sa)->sin_addr;
2122 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
2125 if (cmd == PRC_MSGSIZE)
2126 notify = tcp_mtudisc_notify;
2127 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2128 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2129 cmd == PRC_TIMXCEED_INTRANS) && ip)
2130 notify = tcp_drop_syn_sent;
2133 * Hostdead is ugly because it goes linearly through all PCBs.
2134 * XXX: We never get this from ICMP, otherwise it makes an
2135 * excellent DoS attack on machines with many connections.
2137 else if (cmd == PRC_HOSTDEAD)
2139 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
2143 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
2147 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
2148 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2149 INP_INFO_RLOCK(&V_tcbinfo);
2150 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
2151 th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2152 if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2153 /* signal EHOSTDOWN, as it flushes the cached route */
2154 inp = (*notify)(inp, EHOSTDOWN);
2157 icmp_tcp_seq = th->th_seq;
2159 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2160 !(inp->inp_flags & INP_DROPPED) &&
2161 !(inp->inp_socket == NULL)) {
2162 tp = intotcpcb(inp);
2163 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2164 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2165 if (cmd == PRC_MSGSIZE) {
2168 * If we got a needfrag set the MTU
2169 * in the route to the suggested new
2170 * value (if given) and then notify.
2172 mtu = ntohs(icp->icmp_nextmtu);
2174 * If no alternative MTU was
2175 * proposed, try the next smaller
2180 ntohs(ip->ip_len), 1);
2181 if (mtu < V_tcp_minmss +
2182 sizeof(struct tcpiphdr))
2183 mtu = V_tcp_minmss +
2184 sizeof(struct tcpiphdr);
2186 * Only process the offered MTU if it
2187 * is smaller than the current one.
2189 if (mtu < tp->t_maxseg +
2190 sizeof(struct tcpiphdr)) {
2191 bzero(&inc, sizeof(inc));
2192 inc.inc_faddr = faddr;
2194 inp->inp_inc.inc_fibnum;
2195 tcp_hc_updatemtu(&inc, mtu);
2196 tcp_mtudisc(inp, mtu);
2199 inp = (*notify)(inp,
2200 inetctlerrmap[cmd]);
2204 bzero(&inc, sizeof(inc));
2205 inc.inc_fport = th->th_dport;
2206 inc.inc_lport = th->th_sport;
2207 inc.inc_faddr = faddr;
2208 inc.inc_laddr = ip->ip_src;
2209 syncache_unreach(&inc, icmp_tcp_seq);
2214 INP_INFO_RUNLOCK(&V_tcbinfo);
2220 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
2222 struct in6_addr *dst;
2223 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2224 struct ip6_hdr *ip6;
2228 struct icmp6_hdr *icmp6;
2229 struct ip6ctlparam *ip6cp = NULL;
2230 const struct sockaddr_in6 *sa6_src = NULL;
2231 struct in_conninfo inc;
2236 tcp_seq icmp_tcp_seq;
2240 if (sa->sa_family != AF_INET6 ||
2241 sa->sa_len != sizeof(struct sockaddr_in6))
2244 /* if the parameter is from icmp6, decode it. */
2246 ip6cp = (struct ip6ctlparam *)d;
2247 icmp6 = ip6cp->ip6c_icmp6;
2249 ip6 = ip6cp->ip6c_ip6;
2250 off = ip6cp->ip6c_off;
2251 sa6_src = ip6cp->ip6c_src;
2252 dst = ip6cp->ip6c_finaldst;
2256 off = 0; /* fool gcc */
2261 if (cmd == PRC_MSGSIZE)
2262 notify = tcp_mtudisc_notify;
2263 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2264 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2265 cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
2266 notify = tcp_drop_syn_sent;
2269 * Hostdead is ugly because it goes linearly through all PCBs.
2270 * XXX: We never get this from ICMP, otherwise it makes an
2271 * excellent DoS attack on machines with many connections.
2273 else if (cmd == PRC_HOSTDEAD)
2275 else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
2279 in6_pcbnotify(&V_tcbinfo, sa, 0,
2280 (const struct sockaddr *)sa6_src,
2281 0, cmd, NULL, notify);
2285 /* Check if we can safely get the ports from the tcp hdr */
2288 (int32_t) (off + sizeof(struct tcp_ports)))) {
2291 bzero(&t_ports, sizeof(struct tcp_ports));
2292 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
2293 INP_INFO_RLOCK(&V_tcbinfo);
2294 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
2295 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
2296 if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2297 /* signal EHOSTDOWN, as it flushes the cached route */
2298 inp = (*notify)(inp, EHOSTDOWN);
2301 off += sizeof(struct tcp_ports);
2302 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
2305 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
2307 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2308 !(inp->inp_flags & INP_DROPPED) &&
2309 !(inp->inp_socket == NULL)) {
2310 tp = intotcpcb(inp);
2311 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2312 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2313 if (cmd == PRC_MSGSIZE) {
2316 * If we got a needfrag set the MTU
2317 * in the route to the suggested new
2318 * value (if given) and then notify.
2320 mtu = ntohl(icmp6->icmp6_mtu);
2322 * If no alternative MTU was
2323 * proposed, or the proposed
2324 * MTU was too small, set to
2327 if (mtu < IPV6_MMTU)
2328 mtu = IPV6_MMTU - 8;
2329 bzero(&inc, sizeof(inc));
2330 inc.inc_fibnum = M_GETFIB(m);
2331 inc.inc_flags |= INC_ISIPV6;
2332 inc.inc6_faddr = *dst;
2333 if (in6_setscope(&inc.inc6_faddr,
2334 m->m_pkthdr.rcvif, NULL))
2337 * Only process the offered MTU if it
2338 * is smaller than the current one.
2340 if (mtu < tp->t_maxseg +
2341 sizeof (struct tcphdr) +
2342 sizeof (struct ip6_hdr)) {
2343 tcp_hc_updatemtu(&inc, mtu);
2344 tcp_mtudisc(inp, mtu);
2345 ICMP6STAT_INC(icp6s_pmtuchg);
2348 inp = (*notify)(inp,
2349 inet6ctlerrmap[cmd]);
2353 bzero(&inc, sizeof(inc));
2354 inc.inc_fibnum = M_GETFIB(m);
2355 inc.inc_flags |= INC_ISIPV6;
2356 inc.inc_fport = t_ports.th_dport;
2357 inc.inc_lport = t_ports.th_sport;
2358 inc.inc6_faddr = *dst;
2359 inc.inc6_laddr = ip6->ip6_src;
2360 syncache_unreach(&inc, icmp_tcp_seq);
2365 INP_INFO_RUNLOCK(&V_tcbinfo);
2371 * Following is where TCP initial sequence number generation occurs.
2373 * There are two places where we must use initial sequence numbers:
2374 * 1. In SYN-ACK packets.
2375 * 2. In SYN packets.
2377 * All ISNs for SYN-ACK packets are generated by the syncache. See
2378 * tcp_syncache.c for details.
2380 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2381 * depends on this property. In addition, these ISNs should be
2382 * unguessable so as to prevent connection hijacking. To satisfy
2383 * the requirements of this situation, the algorithm outlined in
2384 * RFC 1948 is used, with only small modifications.
2386 * Implementation details:
2388 * Time is based off the system timer, and is corrected so that it
2389 * increases by one megabyte per second. This allows for proper
2390 * recycling on high speed LANs while still leaving over an hour
2393 * As reading the *exact* system time is too expensive to be done
2394 * whenever setting up a TCP connection, we increment the time
2395 * offset in two ways. First, a small random positive increment
2396 * is added to isn_offset for each connection that is set up.
2397 * Second, the function tcp_isn_tick fires once per clock tick
2398 * and increments isn_offset as necessary so that sequence numbers
2399 * are incremented at approximately ISN_BYTES_PER_SECOND. The
2400 * random positive increments serve only to ensure that the same
2401 * exact sequence number is never sent out twice (as could otherwise
2402 * happen when a port is recycled in less than the system tick
2405 * net.inet.tcp.isn_reseed_interval controls the number of seconds
2406 * between seeding of isn_secret. This is normally set to zero,
2407 * as reseeding should not be necessary.
2409 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2410 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
2411 * general, this means holding an exclusive (write) lock.
2414 #define ISN_BYTES_PER_SECOND 1048576
2415 #define ISN_STATIC_INCREMENT 4096
2416 #define ISN_RANDOM_INCREMENT (4096 - 1)
2418 static VNET_DEFINE(u_char, isn_secret[32]);
2419 static VNET_DEFINE(int, isn_last);
2420 static VNET_DEFINE(int, isn_last_reseed);
2421 static VNET_DEFINE(u_int32_t, isn_offset);
2422 static VNET_DEFINE(u_int32_t, isn_offset_old);
2424 #define V_isn_secret VNET(isn_secret)
2425 #define V_isn_last VNET(isn_last)
2426 #define V_isn_last_reseed VNET(isn_last_reseed)
2427 #define V_isn_offset VNET(isn_offset)
2428 #define V_isn_offset_old VNET(isn_offset_old)
2431 tcp_new_isn(struct tcpcb *tp)
2434 u_int32_t md5_buffer[4];
2436 u_int32_t projected_offset;
2438 INP_WLOCK_ASSERT(tp->t_inpcb);
2441 /* Seed if this is the first use, reseed if requested. */
2442 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2443 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2445 read_random(&V_isn_secret, sizeof(V_isn_secret));
2446 V_isn_last_reseed = ticks;
2449 /* Compute the md5 hash and return the ISN. */
2451 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
2452 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
2454 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
2455 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
2456 sizeof(struct in6_addr));
2457 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
2458 sizeof(struct in6_addr));
2462 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
2463 sizeof(struct in_addr));
2464 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
2465 sizeof(struct in_addr));
2467 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
2468 MD5Final((u_char *) &md5_buffer, &isn_ctx);
2469 new_isn = (tcp_seq) md5_buffer[0];
2470 V_isn_offset += ISN_STATIC_INCREMENT +
2471 (arc4random() & ISN_RANDOM_INCREMENT);
2472 if (ticks != V_isn_last) {
2473 projected_offset = V_isn_offset_old +
2474 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2475 if (SEQ_GT(projected_offset, V_isn_offset))
2476 V_isn_offset = projected_offset;
2477 V_isn_offset_old = V_isn_offset;
2480 new_isn += V_isn_offset;
2486 * When a specific ICMP unreachable message is received and the
2487 * connection state is SYN-SENT, drop the connection. This behavior
2488 * is controlled by the icmp_may_rst sysctl.
2491 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2495 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2496 INP_WLOCK_ASSERT(inp);
2498 if ((inp->inp_flags & INP_TIMEWAIT) ||
2499 (inp->inp_flags & INP_DROPPED))
2502 tp = intotcpcb(inp);
2503 if (tp->t_state != TCPS_SYN_SENT)
2506 if (IS_FASTOPEN(tp->t_flags))
2507 tcp_fastopen_disable_path(tp);
2509 tp = tcp_drop(tp, errno);
2517 * When `need fragmentation' ICMP is received, update our idea of the MSS
2518 * based on the new value. Also nudge TCP to send something, since we
2519 * know the packet we just sent was dropped.
2520 * This duplicates some code in the tcp_mss() function in tcp_input.c.
2522 static struct inpcb *
2523 tcp_mtudisc_notify(struct inpcb *inp, int error)
2526 tcp_mtudisc(inp, -1);
2531 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2536 INP_WLOCK_ASSERT(inp);
2537 if ((inp->inp_flags & INP_TIMEWAIT) ||
2538 (inp->inp_flags & INP_DROPPED))
2541 tp = intotcpcb(inp);
2542 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2544 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2546 so = inp->inp_socket;
2547 SOCKBUF_LOCK(&so->so_snd);
2548 /* If the mss is larger than the socket buffer, decrease the mss. */
2549 if (so->so_snd.sb_hiwat < tp->t_maxseg)
2550 tp->t_maxseg = so->so_snd.sb_hiwat;
2551 SOCKBUF_UNLOCK(&so->so_snd);
2553 TCPSTAT_INC(tcps_mturesent);
2555 tp->snd_nxt = tp->snd_una;
2556 tcp_free_sackholes(tp);
2557 tp->snd_recover = tp->snd_max;
2558 if (tp->t_flags & TF_SACK_PERMIT)
2559 EXIT_FASTRECOVERY(tp->t_flags);
2560 tp->t_fb->tfb_tcp_output(tp);
2565 * Look-up the routing entry to the peer of this inpcb. If no route
2566 * is found and it cannot be allocated, then return 0. This routine
2567 * is called by TCP routines that access the rmx structure and by
2568 * tcp_mss_update to get the peer/interface MTU.
2571 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2573 struct nhop4_extended nh4;
2575 uint32_t maxmtu = 0;
2577 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2579 if (inc->inc_faddr.s_addr != INADDR_ANY) {
2581 if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr,
2582 NHR_REF, 0, &nh4) != 0)
2586 maxmtu = nh4.nh_mtu;
2588 /* Report additional interface capabilities. */
2590 if (ifp->if_capenable & IFCAP_TSO4 &&
2591 ifp->if_hwassist & CSUM_TSO) {
2592 cap->ifcap |= CSUM_TSO;
2593 cap->tsomax = ifp->if_hw_tsomax;
2594 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2595 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2598 fib4_free_nh_ext(inc->inc_fibnum, &nh4);
2606 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2608 struct nhop6_extended nh6;
2609 struct in6_addr dst6;
2612 uint32_t maxmtu = 0;
2614 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2616 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2617 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
2618 if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0,
2623 maxmtu = nh6.nh_mtu;
2625 /* Report additional interface capabilities. */
2627 if (ifp->if_capenable & IFCAP_TSO6 &&
2628 ifp->if_hwassist & CSUM_TSO) {
2629 cap->ifcap |= CSUM_TSO;
2630 cap->tsomax = ifp->if_hw_tsomax;
2631 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2632 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2635 fib6_free_nh_ext(inc->inc_fibnum, &nh6);
2643 * Calculate effective SMSS per RFC5681 definition for a given TCP
2644 * connection at its current state, taking into account SACK and etc.
2647 tcp_maxseg(const struct tcpcb *tp)
2651 if (tp->t_flags & TF_NOOPT)
2652 return (tp->t_maxseg);
2655 * Here we have a simplified code from tcp_addoptions(),
2656 * without a proper loop, and having most of paddings hardcoded.
2657 * We might make mistakes with padding here in some edge cases,
2658 * but this is harmless, since result of tcp_maxseg() is used
2659 * only in cwnd and ssthresh estimations.
2661 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4)
2662 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2663 if (tp->t_flags & TF_RCVD_TSTMP)
2664 optlen = TCPOLEN_TSTAMP_APPA;
2667 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2668 if (tp->t_flags & TF_SIGNATURE)
2669 optlen += PAD(TCPOLEN_SIGNATURE);
2671 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
2672 optlen += TCPOLEN_SACKHDR;
2673 optlen += tp->rcv_numsacks * TCPOLEN_SACK;
2674 optlen = PAD(optlen);
2677 if (tp->t_flags & TF_REQ_TSTMP)
2678 optlen = TCPOLEN_TSTAMP_APPA;
2680 optlen = PAD(TCPOLEN_MAXSEG);
2681 if (tp->t_flags & TF_REQ_SCALE)
2682 optlen += PAD(TCPOLEN_WINDOW);
2683 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2684 if (tp->t_flags & TF_SIGNATURE)
2685 optlen += PAD(TCPOLEN_SIGNATURE);
2687 if (tp->t_flags & TF_SACK_PERMIT)
2688 optlen += PAD(TCPOLEN_SACK_PERMITTED);
2691 optlen = min(optlen, TCP_MAXOLEN);
2692 return (tp->t_maxseg - optlen);
2696 sysctl_drop(SYSCTL_HANDLER_ARGS)
2698 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2699 struct sockaddr_storage addrs[2];
2703 struct sockaddr_in *fin, *lin;
2705 struct sockaddr_in6 *fin6, *lin6;
2716 if (req->oldptr != NULL || req->oldlen != 0)
2718 if (req->newptr == NULL)
2720 if (req->newlen < sizeof(addrs))
2722 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2726 switch (addrs[0].ss_family) {
2729 fin6 = (struct sockaddr_in6 *)&addrs[0];
2730 lin6 = (struct sockaddr_in6 *)&addrs[1];
2731 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2732 lin6->sin6_len != sizeof(struct sockaddr_in6))
2734 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2735 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2737 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2738 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2739 fin = (struct sockaddr_in *)&addrs[0];
2740 lin = (struct sockaddr_in *)&addrs[1];
2743 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2746 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2753 fin = (struct sockaddr_in *)&addrs[0];
2754 lin = (struct sockaddr_in *)&addrs[1];
2755 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2756 lin->sin_len != sizeof(struct sockaddr_in))
2763 INP_INFO_RLOCK(&V_tcbinfo);
2764 switch (addrs[0].ss_family) {
2767 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2768 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2769 INPLOOKUP_WLOCKPCB, NULL);
2774 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2775 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2780 if (inp->inp_flags & INP_TIMEWAIT) {
2782 * XXXRW: There currently exists a state where an
2783 * inpcb is present, but its timewait state has been
2784 * discarded. For now, don't allow dropping of this
2792 } else if (!(inp->inp_flags & INP_DROPPED) &&
2793 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2794 tp = intotcpcb(inp);
2795 tp = tcp_drop(tp, ECONNABORTED);
2802 INP_INFO_RUNLOCK(&V_tcbinfo);
2806 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2807 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
2808 0, sysctl_drop, "", "Drop TCP connection");
2811 * Generate a standardized TCP log line for use throughout the
2812 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2813 * allow use in the interrupt context.
2815 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2816 * NB: The function may return NULL if memory allocation failed.
2818 * Due to header inclusion and ordering limitations the struct ip
2819 * and ip6_hdr pointers have to be passed as void pointers.
2822 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2826 /* Is logging enabled? */
2827 if (tcp_log_in_vain == 0)
2830 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2834 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2838 /* Is logging enabled? */
2839 if (tcp_log_debug == 0)
2842 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2846 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2853 const struct ip6_hdr *ip6;
2855 ip6 = (const struct ip6_hdr *)ip6hdr;
2857 ip = (struct ip *)ip4hdr;
2860 * The log line looks like this:
2861 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2863 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2864 sizeof(PRINT_TH_FLAGS) + 1 +
2866 2 * INET6_ADDRSTRLEN;
2868 2 * INET_ADDRSTRLEN;
2871 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2875 strcat(s, "TCP: [");
2878 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2879 inet_ntoa_r(inc->inc_faddr, sp);
2881 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2883 inet_ntoa_r(inc->inc_laddr, sp);
2885 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2888 ip6_sprintf(sp, &inc->inc6_faddr);
2890 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2892 ip6_sprintf(sp, &inc->inc6_laddr);
2894 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2895 } else if (ip6 && th) {
2896 ip6_sprintf(sp, &ip6->ip6_src);
2898 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2900 ip6_sprintf(sp, &ip6->ip6_dst);
2902 sprintf(sp, "]:%i", ntohs(th->th_dport));
2905 } else if (ip && th) {
2906 inet_ntoa_r(ip->ip_src, sp);
2908 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2910 inet_ntoa_r(ip->ip_dst, sp);
2912 sprintf(sp, "]:%i", ntohs(th->th_dport));
2920 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2921 if (*(s + size - 1) != '\0')
2922 panic("%s: string too long", __func__);
2927 * A subroutine which makes it easy to track TCP state changes with DTrace.
2928 * This function shouldn't be called for t_state initializations that don't
2929 * correspond to actual TCP state transitions.
2932 tcp_state_change(struct tcpcb *tp, int newstate)
2934 #if defined(KDTRACE_HOOKS)
2935 int pstate = tp->t_state;
2938 TCPSTATES_DEC(tp->t_state);
2939 TCPSTATES_INC(newstate);
2940 tp->t_state = newstate;
2941 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
2945 * Create an external-format (``xtcpcb'') structure using the information in
2946 * the kernel-format tcpcb structure pointed to by tp. This is done to
2947 * reduce the spew of irrelevant information over this interface, to isolate
2948 * user code from changes in the kernel structure, and potentially to provide
2949 * information-hiding if we decide that some of this information should be
2950 * hidden from users.
2953 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
2955 struct tcpcb *tp = intotcpcb(inp);
2958 if (inp->inp_flags & INP_TIMEWAIT) {
2959 bzero(xt, sizeof(struct xtcpcb));
2960 xt->t_state = TCPS_TIME_WAIT;
2962 xt->t_state = tp->t_state;
2963 xt->t_logstate = tp->t_logstate;
2964 xt->t_flags = tp->t_flags;
2965 xt->t_sndzerowin = tp->t_sndzerowin;
2966 xt->t_sndrexmitpack = tp->t_sndrexmitpack;
2967 xt->t_rcvoopack = tp->t_rcvoopack;
2969 now = getsbinuptime();
2970 #define COPYTIMER(ttt) do { \
2971 if (callout_active(&tp->t_timers->ttt)) \
2972 xt->ttt = (tp->t_timers->ttt.c_time - now) / \
2977 COPYTIMER(tt_delack);
2978 COPYTIMER(tt_rexmt);
2979 COPYTIMER(tt_persist);
2983 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
2985 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
2986 TCP_FUNCTION_NAME_LEN_MAX);
2987 bzero(xt->xt_logid, TCP_LOG_ID_LEN);
2989 (void)tcp_log_get_id(tp, xt->xt_logid);
2993 xt->xt_len = sizeof(struct xtcpcb);
2994 in_pcbtoxinpcb(inp, &xt->xt_inp);
2995 if (inp->inp_socket == NULL)
2996 xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP;