1 /*-
2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3 * The Regents of the University of California. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 *
29 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95
30 */
31
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34
35 #include "opt_compat.h"
36 #include "opt_inet.h"
37 #include "opt_inet6.h"
38 #include "opt_ipsec.h"
39 #include "opt_tcpdebug.h"
40
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/callout.h>
44 #include <sys/eventhandler.h>
45 #include <sys/hhook.h>
46 #include <sys/kernel.h>
47 #include <sys/khelp.h>
48 #include <sys/sysctl.h>
49 #include <sys/jail.h>
50 #include <sys/malloc.h>
51 #include <sys/refcount.h>
52 #include <sys/mbuf.h>
53 #ifdef INET6
54 #include <sys/domain.h>
55 #endif
56 #include <sys/priv.h>
57 #include <sys/proc.h>
58 #include <sys/sdt.h>
59 #include <sys/socket.h>
60 #include <sys/socketvar.h>
61 #include <sys/protosw.h>
62 #include <sys/random.h>
63
64 #include <vm/uma.h>
65
66 #include <net/route.h>
67 #include <net/if.h>
68 #include <net/if_var.h>
69 #include <net/vnet.h>
70
71 #include <netinet/in.h>
72 #include <netinet/in_fib.h>
73 #include <netinet/in_kdtrace.h>
74 #include <netinet/in_pcb.h>
75 #include <netinet/in_systm.h>
76 #include <netinet/in_var.h>
77 #include <netinet/ip.h>
78 #include <netinet/ip_icmp.h>
79 #include <netinet/ip_var.h>
80 #ifdef INET6
81 #include <netinet/icmp6.h>
82 #include <netinet/ip6.h>
83 #include <netinet6/in6_fib.h>
84 #include <netinet6/in6_pcb.h>
85 #include <netinet6/ip6_var.h>
86 #include <netinet6/scope6_var.h>
87 #include <netinet6/nd6.h>
88 #endif
89
90 #ifdef TCP_RFC7413
91 #include <netinet/tcp_fastopen.h>
92 #endif
93 #include <netinet/tcp.h>
94 #include <netinet/tcp_fsm.h>
95 #include <netinet/tcp_seq.h>
96 #include <netinet/tcp_timer.h>
97 #include <netinet/tcp_var.h>
98 #include <netinet/tcp_syncache.h>
99 #include <netinet/cc/cc.h>
100 #ifdef INET6
101 #include <netinet6/tcp6_var.h>
102 #endif
103 #include <netinet/tcpip.h>
104 #ifdef TCPPCAP
105 #include <netinet/tcp_pcap.h>
106 #endif
107 #ifdef TCPDEBUG
108 #include <netinet/tcp_debug.h>
109 #endif
110 #ifdef INET6
111 #include <netinet6/ip6protosw.h>
112 #endif
113 #ifdef TCP_OFFLOAD
114 #include <netinet/tcp_offload.h>
115 #endif
116
117 #include <netipsec/ipsec_support.h>
118
119 #include <machine/in_cksum.h>
120 #include <sys/md5.h>
121
122 #include <security/mac/mac_framework.h>
123
124 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
125 #ifdef INET6
126 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
127 #endif
128
129 struct rwlock tcp_function_lock;
130
131 static int
sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)132 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
133 {
134 int error, new;
135
136 new = V_tcp_mssdflt;
137 error = sysctl_handle_int(oidp, &new, 0, req);
138 if (error == 0 && req->newptr) {
139 if (new < TCP_MINMSS)
140 error = EINVAL;
141 else
142 V_tcp_mssdflt = new;
143 }
144 return (error);
145 }
146
147 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
148 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
149 &sysctl_net_inet_tcp_mss_check, "I",
150 "Default TCP Maximum Segment Size");
151
152 #ifdef INET6
153 static int
sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)154 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
155 {
156 int error, new;
157
158 new = V_tcp_v6mssdflt;
159 error = sysctl_handle_int(oidp, &new, 0, req);
160 if (error == 0 && req->newptr) {
161 if (new < TCP_MINMSS)
162 error = EINVAL;
163 else
164 V_tcp_v6mssdflt = new;
165 }
166 return (error);
167 }
168
169 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
170 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
171 &sysctl_net_inet_tcp_mss_v6_check, "I",
172 "Default TCP Maximum Segment Size for IPv6");
173 #endif /* INET6 */
174
175 /*
176 * Minimum MSS we accept and use. This prevents DoS attacks where
177 * we are forced to a ridiculous low MSS like 20 and send hundreds
178 * of packets instead of one. The effect scales with the available
179 * bandwidth and quickly saturates the CPU and network interface
180 * with packet generation and sending. Set to zero to disable MINMSS
181 * checking. This setting prevents us from sending too small packets.
182 */
183 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
184 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
185 &VNET_NAME(tcp_minmss), 0,
186 "Minimum TCP Maximum Segment Size");
187
188 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
189 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
190 &VNET_NAME(tcp_do_rfc1323), 0,
191 "Enable rfc1323 (high performance TCP) extensions");
192
193 static int tcp_log_debug = 0;
194 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
195 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
196
197 static int tcp_tcbhashsize;
198 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
199 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
200
201 static int do_tcpdrain = 1;
202 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
203 "Enable tcp_drain routine for extra help when low on mbufs");
204
205 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
206 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
207
208 static VNET_DEFINE(int, icmp_may_rst) = 1;
209 #define V_icmp_may_rst VNET(icmp_may_rst)
210 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
211 &VNET_NAME(icmp_may_rst), 0,
212 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
213
214 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
215 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
216 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
217 &VNET_NAME(tcp_isn_reseed_interval), 0,
218 "Seconds between reseeding of ISN secret");
219
220 static int tcp_soreceive_stream;
221 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
222 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
223
224 VNET_DEFINE(uma_zone_t, sack_hole_zone);
225 #define V_sack_hole_zone VNET(sack_hole_zone)
226
227 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
228
229 VNET_DEFINE(u_char, ts_offset_secret[32]);
230 #define V_ts_offset_secret VNET(ts_offset_secret)
231
232 static struct inpcb *tcp_notify(struct inpcb *, int);
233 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
234 static void tcp_mtudisc(struct inpcb *, int);
235 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
236 void *ip4hdr, const void *ip6hdr);
237
238
239 static struct tcp_function_block tcp_def_funcblk = {
240 "default",
241 tcp_output,
242 tcp_do_segment,
243 tcp_default_ctloutput,
244 NULL,
245 NULL,
246 NULL,
247 NULL,
248 NULL,
249 NULL,
250 0,
251 0
252 };
253
254 int t_functions_inited = 0;
255 struct tcp_funchead t_functions;
256 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
257
258 static void
init_tcp_functions(void)259 init_tcp_functions(void)
260 {
261 if (t_functions_inited == 0) {
262 TAILQ_INIT(&t_functions);
263 rw_init_flags(&tcp_function_lock, "tcp_func_lock" , 0);
264 t_functions_inited = 1;
265 }
266 }
267
268 static struct tcp_function_block *
find_tcp_functions_locked(struct tcp_function_set * fs)269 find_tcp_functions_locked(struct tcp_function_set *fs)
270 {
271 struct tcp_function *f;
272 struct tcp_function_block *blk=NULL;
273
274 TAILQ_FOREACH(f, &t_functions, tf_next) {
275 if (strcmp(f->tf_fb->tfb_tcp_block_name, fs->function_set_name) == 0) {
276 blk = f->tf_fb;
277 break;
278 }
279 }
280 return(blk);
281 }
282
283 static struct tcp_function_block *
find_tcp_fb_locked(struct tcp_function_block * blk,struct tcp_function ** s)284 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
285 {
286 struct tcp_function_block *rblk=NULL;
287 struct tcp_function *f;
288
289 TAILQ_FOREACH(f, &t_functions, tf_next) {
290 if (f->tf_fb == blk) {
291 rblk = blk;
292 if (s) {
293 *s = f;
294 }
295 break;
296 }
297 }
298 return (rblk);
299 }
300
301 struct tcp_function_block *
find_and_ref_tcp_functions(struct tcp_function_set * fs)302 find_and_ref_tcp_functions(struct tcp_function_set *fs)
303 {
304 struct tcp_function_block *blk;
305
306 rw_rlock(&tcp_function_lock);
307 blk = find_tcp_functions_locked(fs);
308 if (blk)
309 refcount_acquire(&blk->tfb_refcnt);
310 rw_runlock(&tcp_function_lock);
311 return(blk);
312 }
313
314 struct tcp_function_block *
find_and_ref_tcp_fb(struct tcp_function_block * blk)315 find_and_ref_tcp_fb(struct tcp_function_block *blk)
316 {
317 struct tcp_function_block *rblk;
318
319 rw_rlock(&tcp_function_lock);
320 rblk = find_tcp_fb_locked(blk, NULL);
321 if (rblk)
322 refcount_acquire(&rblk->tfb_refcnt);
323 rw_runlock(&tcp_function_lock);
324 return(rblk);
325 }
326
327
328 static int
sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)329 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
330 {
331 int error=ENOENT;
332 struct tcp_function_set fs;
333 struct tcp_function_block *blk;
334
335 memset(&fs, 0, sizeof(fs));
336 rw_rlock(&tcp_function_lock);
337 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
338 if (blk) {
339 /* Found him */
340 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
341 fs.pcbcnt = blk->tfb_refcnt;
342 }
343 rw_runlock(&tcp_function_lock);
344 error = sysctl_handle_string(oidp, fs.function_set_name,
345 sizeof(fs.function_set_name), req);
346
347 /* Check for error or no change */
348 if (error != 0 || req->newptr == NULL)
349 return(error);
350
351 rw_wlock(&tcp_function_lock);
352 blk = find_tcp_functions_locked(&fs);
353 if ((blk == NULL) ||
354 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
355 error = ENOENT;
356 goto done;
357 }
358 tcp_func_set_ptr = blk;
359 done:
360 rw_wunlock(&tcp_function_lock);
361 return (error);
362 }
363
364 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
365 CTLTYPE_STRING | CTLFLAG_RW,
366 NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
367 "Set/get the default TCP functions");
368
369 static int
sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)370 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
371 {
372 int error, cnt, linesz;
373 struct tcp_function *f;
374 char *buffer, *cp;
375 size_t bufsz, outsz;
376
377 cnt = 0;
378 rw_rlock(&tcp_function_lock);
379 TAILQ_FOREACH(f, &t_functions, tf_next) {
380 cnt++;
381 }
382 rw_runlock(&tcp_function_lock);
383
384 bufsz = (cnt+2) * (TCP_FUNCTION_NAME_LEN_MAX + 12) + 1;
385 buffer = malloc(bufsz, M_TEMP, M_WAITOK);
386
387 error = 0;
388 cp = buffer;
389
390 linesz = snprintf(cp, bufsz, "\n%-32s%c %s\n", "Stack", 'D', "PCB count");
391 cp += linesz;
392 bufsz -= linesz;
393 outsz = linesz;
394
395 rw_rlock(&tcp_function_lock);
396 TAILQ_FOREACH(f, &t_functions, tf_next) {
397 linesz = snprintf(cp, bufsz, "%-32s%c %u\n",
398 f->tf_fb->tfb_tcp_block_name,
399 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
400 f->tf_fb->tfb_refcnt);
401 if (linesz >= bufsz) {
402 error = EOVERFLOW;
403 break;
404 }
405 cp += linesz;
406 bufsz -= linesz;
407 outsz += linesz;
408 }
409 rw_runlock(&tcp_function_lock);
410 if (error == 0)
411 error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
412 free(buffer, M_TEMP);
413 return (error);
414 }
415
416 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
417 CTLTYPE_STRING|CTLFLAG_RD,
418 NULL, 0, sysctl_net_inet_list_available, "A",
419 "list available TCP Function sets");
420
421 /*
422 * Target size of TCP PCB hash tables. Must be a power of two.
423 *
424 * Note that this can be overridden by the kernel environment
425 * variable net.inet.tcp.tcbhashsize
426 */
427 #ifndef TCBHASHSIZE
428 #define TCBHASHSIZE 0
429 #endif
430
431 /*
432 * XXX
433 * Callouts should be moved into struct tcp directly. They are currently
434 * separate because the tcpcb structure is exported to userland for sysctl
435 * parsing purposes, which do not know about callouts.
436 */
437 struct tcpcb_mem {
438 struct tcpcb tcb;
439 struct tcp_timer tt;
440 struct cc_var ccv;
441 struct osd osd;
442 };
443
444 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
445 #define V_tcpcb_zone VNET(tcpcb_zone)
446
447 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
448 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
449
450 static struct mtx isn_mtx;
451
452 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
453 #define ISN_LOCK() mtx_lock(&isn_mtx)
454 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
455
456 /*
457 * TCP initialization.
458 */
459 static void
tcp_zone_change(void * tag)460 tcp_zone_change(void *tag)
461 {
462
463 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
464 uma_zone_set_max(V_tcpcb_zone, maxsockets);
465 tcp_tw_zone_change();
466 }
467
468 static int
tcp_inpcb_init(void * mem,int size,int flags)469 tcp_inpcb_init(void *mem, int size, int flags)
470 {
471 struct inpcb *inp = mem;
472
473 INP_LOCK_INIT(inp, "inp", "tcpinp");
474 return (0);
475 }
476
477 /*
478 * Take a value and get the next power of 2 that doesn't overflow.
479 * Used to size the tcp_inpcb hash buckets.
480 */
481 static int
maketcp_hashsize(int size)482 maketcp_hashsize(int size)
483 {
484 int hashsize;
485
486 /*
487 * auto tune.
488 * get the next power of 2 higher than maxsockets.
489 */
490 hashsize = 1 << fls(size);
491 /* catch overflow, and just go one power of 2 smaller */
492 if (hashsize < size) {
493 hashsize = 1 << (fls(size) - 1);
494 }
495 return (hashsize);
496 }
497
498 int
register_tcp_functions(struct tcp_function_block * blk,int wait)499 register_tcp_functions(struct tcp_function_block *blk, int wait)
500 {
501 struct tcp_function_block *lblk;
502 struct tcp_function *n;
503 struct tcp_function_set fs;
504
505 if (t_functions_inited == 0) {
506 init_tcp_functions();
507 }
508 if ((blk->tfb_tcp_output == NULL) ||
509 (blk->tfb_tcp_do_segment == NULL) ||
510 (blk->tfb_tcp_ctloutput == NULL) ||
511 (strlen(blk->tfb_tcp_block_name) == 0)) {
512 /*
513 * These functions are required and you
514 * need a name.
515 */
516 return (EINVAL);
517 }
518 if (blk->tfb_tcp_timer_stop_all ||
519 blk->tfb_tcp_timer_activate ||
520 blk->tfb_tcp_timer_active ||
521 blk->tfb_tcp_timer_stop) {
522 /*
523 * If you define one timer function you
524 * must have them all.
525 */
526 if ((blk->tfb_tcp_timer_stop_all == NULL) ||
527 (blk->tfb_tcp_timer_activate == NULL) ||
528 (blk->tfb_tcp_timer_active == NULL) ||
529 (blk->tfb_tcp_timer_stop == NULL)) {
530 return (EINVAL);
531 }
532 }
533 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
534 return (EINVAL);
535 }
536 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
537 if (n == NULL) {
538 return (ENOMEM);
539 }
540 n->tf_fb = blk;
541 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
542 rw_wlock(&tcp_function_lock);
543 lblk = find_tcp_functions_locked(&fs);
544 if (lblk) {
545 /* Duplicate name space not allowed */
546 rw_wunlock(&tcp_function_lock);
547 free(n, M_TCPFUNCTIONS);
548 return (EALREADY);
549 }
550 refcount_init(&blk->tfb_refcnt, 0);
551 TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
552 rw_wunlock(&tcp_function_lock);
553 return(0);
554 }
555
556 int
deregister_tcp_functions(struct tcp_function_block * blk)557 deregister_tcp_functions(struct tcp_function_block *blk)
558 {
559 struct tcp_function_block *lblk;
560 struct tcp_function *f;
561 int error=ENOENT;
562
563 if (strcmp(blk->tfb_tcp_block_name, "default") == 0) {
564 /* You can't un-register the default */
565 return (EPERM);
566 }
567 rw_wlock(&tcp_function_lock);
568 if (blk == tcp_func_set_ptr) {
569 /* You can't free the current default */
570 rw_wunlock(&tcp_function_lock);
571 return (EBUSY);
572 }
573 if (blk->tfb_refcnt) {
574 /* Still tcb attached, mark it. */
575 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
576 rw_wunlock(&tcp_function_lock);
577 return (EBUSY);
578 }
579 lblk = find_tcp_fb_locked(blk, &f);
580 if (lblk) {
581 /* Found */
582 TAILQ_REMOVE(&t_functions, f, tf_next);
583 f->tf_fb = NULL;
584 free(f, M_TCPFUNCTIONS);
585 error = 0;
586 }
587 rw_wunlock(&tcp_function_lock);
588 return (error);
589 }
590
591 void
tcp_init(void)592 tcp_init(void)
593 {
594 const char *tcbhash_tuneable;
595 int hashsize;
596
597 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
598
599 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
600 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
601 printf("%s: WARNING: unable to register helper hook\n", __func__);
602 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
603 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
604 printf("%s: WARNING: unable to register helper hook\n", __func__);
605 hashsize = TCBHASHSIZE;
606 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
607 if (hashsize == 0) {
608 /*
609 * Auto tune the hash size based on maxsockets.
610 * A perfect hash would have a 1:1 mapping
611 * (hashsize = maxsockets) however it's been
612 * suggested that O(2) average is better.
613 */
614 hashsize = maketcp_hashsize(maxsockets / 4);
615 /*
616 * Our historical default is 512,
617 * do not autotune lower than this.
618 */
619 if (hashsize < 512)
620 hashsize = 512;
621 if (bootverbose && IS_DEFAULT_VNET(curvnet))
622 printf("%s: %s auto tuned to %d\n", __func__,
623 tcbhash_tuneable, hashsize);
624 }
625 /*
626 * We require a hashsize to be a power of two.
627 * Previously if it was not a power of two we would just reset it
628 * back to 512, which could be a nasty surprise if you did not notice
629 * the error message.
630 * Instead what we do is clip it to the closest power of two lower
631 * than the specified hash value.
632 */
633 if (!powerof2(hashsize)) {
634 int oldhashsize = hashsize;
635
636 hashsize = maketcp_hashsize(hashsize);
637 /* prevent absurdly low value */
638 if (hashsize < 16)
639 hashsize = 16;
640 printf("%s: WARNING: TCB hash size not a power of 2, "
641 "clipped from %d to %d.\n", __func__, oldhashsize,
642 hashsize);
643 }
644 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
645 "tcp_inpcb", tcp_inpcb_init, NULL, 0, IPI_HASHFIELDS_4TUPLE);
646
647 /*
648 * These have to be type stable for the benefit of the timers.
649 */
650 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
651 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
652 uma_zone_set_max(V_tcpcb_zone, maxsockets);
653 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
654
655 tcp_tw_init();
656 syncache_init();
657 tcp_hc_init();
658
659 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
660 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
661 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
662
663 #ifdef TCP_RFC7413
664 tcp_fastopen_init();
665 #endif
666
667 /* Skip initialization of globals for non-default instances. */
668 if (!IS_DEFAULT_VNET(curvnet))
669 return;
670
671 tcp_reass_global_init();
672
673 /* XXX virtualize those bellow? */
674 tcp_delacktime = TCPTV_DELACK;
675 tcp_keepinit = TCPTV_KEEP_INIT;
676 tcp_keepidle = TCPTV_KEEP_IDLE;
677 tcp_keepintvl = TCPTV_KEEPINTVL;
678 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
679 tcp_msl = TCPTV_MSL;
680 tcp_rexmit_min = TCPTV_MIN;
681 if (tcp_rexmit_min < 1)
682 tcp_rexmit_min = 1;
683 tcp_persmin = TCPTV_PERSMIN;
684 tcp_persmax = TCPTV_PERSMAX;
685 tcp_rexmit_slop = TCPTV_CPU_VAR;
686 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
687 tcp_tcbhashsize = hashsize;
688 /* Setup the tcp function block list */
689 init_tcp_functions();
690 register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
691 read_random(&V_ts_offset_secret, sizeof(V_ts_offset_secret));
692
693 if (tcp_soreceive_stream) {
694 #ifdef INET
695 tcp_usrreqs.pru_soreceive = soreceive_stream;
696 #endif
697 #ifdef INET6
698 tcp6_usrreqs.pru_soreceive = soreceive_stream;
699 #endif /* INET6 */
700 }
701
702 #ifdef INET6
703 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
704 #else /* INET6 */
705 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
706 #endif /* INET6 */
707 if (max_protohdr < TCP_MINPROTOHDR)
708 max_protohdr = TCP_MINPROTOHDR;
709 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
710 panic("tcp_init");
711 #undef TCP_MINPROTOHDR
712
713 ISN_LOCK_INIT();
714 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
715 SHUTDOWN_PRI_DEFAULT);
716 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
717 EVENTHANDLER_PRI_ANY);
718 #ifdef TCPPCAP
719 tcp_pcap_init();
720 #endif
721 }
722
723 #ifdef VIMAGE
724 static void
tcp_destroy(void * unused __unused)725 tcp_destroy(void *unused __unused)
726 {
727 int error, n;
728
729 /*
730 * All our processes are gone, all our sockets should be cleaned
731 * up, which means, we should be past the tcp_discardcb() calls.
732 * Sleep to let all tcpcb timers really disappear and cleanup.
733 */
734 for (;;) {
735 INP_LIST_RLOCK(&V_tcbinfo);
736 n = V_tcbinfo.ipi_count;
737 INP_LIST_RUNLOCK(&V_tcbinfo);
738 if (n == 0)
739 break;
740 pause("tcpdes", hz / 10);
741 }
742 tcp_hc_destroy();
743 syncache_destroy();
744 tcp_tw_destroy();
745 in_pcbinfo_destroy(&V_tcbinfo);
746 /* tcp_discardcb() clears the sack_holes up. */
747 uma_zdestroy(V_sack_hole_zone);
748 uma_zdestroy(V_tcpcb_zone);
749
750 #ifdef TCP_RFC7413
751 /*
752 * Cannot free the zone until all tcpcbs are released as we attach
753 * the allocations to them.
754 */
755 tcp_fastopen_destroy();
756 #endif
757
758 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
759 if (error != 0) {
760 printf("%s: WARNING: unable to deregister helper hook "
761 "type=%d, id=%d: error %d returned\n", __func__,
762 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
763 }
764 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
765 if (error != 0) {
766 printf("%s: WARNING: unable to deregister helper hook "
767 "type=%d, id=%d: error %d returned\n", __func__,
768 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
769 }
770 }
771 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
772 #endif
773
774 void
tcp_fini(void * xtp)775 tcp_fini(void *xtp)
776 {
777
778 }
779
780 /*
781 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
782 * tcp_template used to store this data in mbufs, but we now recopy it out
783 * of the tcpcb each time to conserve mbufs.
784 */
785 void
tcpip_fillheaders(struct inpcb * inp,void * ip_ptr,void * tcp_ptr)786 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
787 {
788 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
789
790 INP_WLOCK_ASSERT(inp);
791
792 #ifdef INET6
793 if ((inp->inp_vflag & INP_IPV6) != 0) {
794 struct ip6_hdr *ip6;
795
796 ip6 = (struct ip6_hdr *)ip_ptr;
797 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
798 (inp->inp_flow & IPV6_FLOWINFO_MASK);
799 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
800 (IPV6_VERSION & IPV6_VERSION_MASK);
801 ip6->ip6_nxt = IPPROTO_TCP;
802 ip6->ip6_plen = htons(sizeof(struct tcphdr));
803 ip6->ip6_src = inp->in6p_laddr;
804 ip6->ip6_dst = inp->in6p_faddr;
805 }
806 #endif /* INET6 */
807 #if defined(INET6) && defined(INET)
808 else
809 #endif
810 #ifdef INET
811 {
812 struct ip *ip;
813
814 ip = (struct ip *)ip_ptr;
815 ip->ip_v = IPVERSION;
816 ip->ip_hl = 5;
817 ip->ip_tos = inp->inp_ip_tos;
818 ip->ip_len = 0;
819 ip->ip_id = 0;
820 ip->ip_off = 0;
821 ip->ip_ttl = inp->inp_ip_ttl;
822 ip->ip_sum = 0;
823 ip->ip_p = IPPROTO_TCP;
824 ip->ip_src = inp->inp_laddr;
825 ip->ip_dst = inp->inp_faddr;
826 }
827 #endif /* INET */
828 th->th_sport = inp->inp_lport;
829 th->th_dport = inp->inp_fport;
830 th->th_seq = 0;
831 th->th_ack = 0;
832 th->th_x2 = 0;
833 th->th_off = 5;
834 th->th_flags = 0;
835 th->th_win = 0;
836 th->th_urp = 0;
837 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
838 }
839
840 /*
841 * Create template to be used to send tcp packets on a connection.
842 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
843 * use for this function is in keepalives, which use tcp_respond.
844 */
845 struct tcptemp *
tcpip_maketemplate(struct inpcb * inp)846 tcpip_maketemplate(struct inpcb *inp)
847 {
848 struct tcptemp *t;
849
850 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
851 if (t == NULL)
852 return (NULL);
853 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
854 return (t);
855 }
856
857 /*
858 * Send a single message to the TCP at address specified by
859 * the given TCP/IP header. If m == NULL, then we make a copy
860 * of the tcpiphdr at th and send directly to the addressed host.
861 * This is used to force keep alive messages out using the TCP
862 * template for a connection. If flags are given then we send
863 * a message back to the TCP which originated the segment th,
864 * and discard the mbuf containing it and any other attached mbufs.
865 *
866 * In any case the ack and sequence number of the transmitted
867 * segment are as specified by the parameters.
868 *
869 * NOTE: If m != NULL, then th must point to *inside* the mbuf.
870 */
871 void
tcp_respond(struct tcpcb * tp,void * ipgen,struct tcphdr * th,struct mbuf * m,tcp_seq ack,tcp_seq seq,int flags)872 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
873 tcp_seq ack, tcp_seq seq, int flags)
874 {
875 struct tcpopt to;
876 struct inpcb *inp;
877 struct ip *ip;
878 struct mbuf *optm;
879 struct tcphdr *nth;
880 u_char *optp;
881 #ifdef INET6
882 struct ip6_hdr *ip6;
883 int isipv6;
884 #endif /* INET6 */
885 int optlen, tlen, win;
886 bool incl_opts;
887
888 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
889
890 #ifdef INET6
891 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
892 ip6 = ipgen;
893 #endif /* INET6 */
894 ip = ipgen;
895
896 if (tp != NULL) {
897 inp = tp->t_inpcb;
898 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
899 INP_WLOCK_ASSERT(inp);
900 } else
901 inp = NULL;
902
903 incl_opts = false;
904 win = 0;
905 if (tp != NULL) {
906 if (!(flags & TH_RST)) {
907 win = sbspace(&inp->inp_socket->so_rcv);
908 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
909 win = (long)TCP_MAXWIN << tp->rcv_scale;
910 }
911 if ((tp->t_flags & TF_NOOPT) == 0)
912 incl_opts = true;
913 }
914 if (m == NULL) {
915 m = m_gethdr(M_NOWAIT, MT_DATA);
916 if (m == NULL)
917 return;
918 m->m_data += max_linkhdr;
919 #ifdef INET6
920 if (isipv6) {
921 bcopy((caddr_t)ip6, mtod(m, caddr_t),
922 sizeof(struct ip6_hdr));
923 ip6 = mtod(m, struct ip6_hdr *);
924 nth = (struct tcphdr *)(ip6 + 1);
925 } else
926 #endif /* INET6 */
927 {
928 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
929 ip = mtod(m, struct ip *);
930 nth = (struct tcphdr *)(ip + 1);
931 }
932 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
933 flags = TH_ACK;
934 } else if (!M_WRITABLE(m)) {
935 struct mbuf *n;
936
937 /* Can't reuse 'm', allocate a new mbuf. */
938 n = m_gethdr(M_NOWAIT, MT_DATA);
939 if (n == NULL) {
940 m_freem(m);
941 return;
942 }
943
944 if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
945 m_freem(m);
946 m_freem(n);
947 return;
948 }
949
950 n->m_data += max_linkhdr;
951 /* m_len is set later */
952 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
953 #ifdef INET6
954 if (isipv6) {
955 bcopy((caddr_t)ip6, mtod(n, caddr_t),
956 sizeof(struct ip6_hdr));
957 ip6 = mtod(n, struct ip6_hdr *);
958 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
959 nth = (struct tcphdr *)(ip6 + 1);
960 } else
961 #endif /* INET6 */
962 {
963 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
964 ip = mtod(n, struct ip *);
965 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
966 nth = (struct tcphdr *)(ip + 1);
967 }
968 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
969 xchg(nth->th_dport, nth->th_sport, uint16_t);
970 th = nth;
971 m_freem(m);
972 m = n;
973 } else {
974 /*
975 * reuse the mbuf.
976 * XXX MRT We inherit the FIB, which is lucky.
977 */
978 m_freem(m->m_next);
979 m->m_next = NULL;
980 m->m_data = (caddr_t)ipgen;
981 /* m_len is set later */
982 #ifdef INET6
983 if (isipv6) {
984 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
985 nth = (struct tcphdr *)(ip6 + 1);
986 } else
987 #endif /* INET6 */
988 {
989 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
990 nth = (struct tcphdr *)(ip + 1);
991 }
992 if (th != nth) {
993 /*
994 * this is usually a case when an extension header
995 * exists between the IPv6 header and the
996 * TCP header.
997 */
998 nth->th_sport = th->th_sport;
999 nth->th_dport = th->th_dport;
1000 }
1001 xchg(nth->th_dport, nth->th_sport, uint16_t);
1002 #undef xchg
1003 }
1004 tlen = 0;
1005 #ifdef INET6
1006 if (isipv6)
1007 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1008 #endif
1009 #if defined(INET) && defined(INET6)
1010 else
1011 #endif
1012 #ifdef INET
1013 tlen = sizeof (struct tcpiphdr);
1014 #endif
1015 #ifdef INVARIANTS
1016 m->m_len = 0;
1017 KASSERT(M_TRAILINGSPACE(m) >= tlen,
1018 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1019 m, tlen, (long)M_TRAILINGSPACE(m)));
1020 #endif
1021 m->m_len = tlen;
1022 to.to_flags = 0;
1023 if (incl_opts) {
1024 /* Make sure we have room. */
1025 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1026 m->m_next = m_get(M_NOWAIT, MT_DATA);
1027 if (m->m_next) {
1028 optp = mtod(m->m_next, u_char *);
1029 optm = m->m_next;
1030 } else
1031 incl_opts = false;
1032 } else {
1033 optp = (u_char *) (nth + 1);
1034 optm = m;
1035 }
1036 }
1037 if (incl_opts) {
1038 /* Timestamps. */
1039 if (tp->t_flags & TF_RCVD_TSTMP) {
1040 to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1041 to.to_tsecr = tp->ts_recent;
1042 to.to_flags |= TOF_TS;
1043 }
1044 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1045 /* TCP-MD5 (RFC2385). */
1046 if (tp->t_flags & TF_SIGNATURE)
1047 to.to_flags |= TOF_SIGNATURE;
1048 #endif
1049 /* Add the options. */
1050 tlen += optlen = tcp_addoptions(&to, optp);
1051
1052 /* Update m_len in the correct mbuf. */
1053 optm->m_len += optlen;
1054 } else
1055 optlen = 0;
1056 #ifdef INET6
1057 if (isipv6) {
1058 ip6->ip6_flow = 0;
1059 ip6->ip6_vfc = IPV6_VERSION;
1060 ip6->ip6_nxt = IPPROTO_TCP;
1061 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1062 }
1063 #endif
1064 #if defined(INET) && defined(INET6)
1065 else
1066 #endif
1067 #ifdef INET
1068 {
1069 ip->ip_len = htons(tlen);
1070 ip->ip_ttl = V_ip_defttl;
1071 if (V_path_mtu_discovery)
1072 ip->ip_off |= htons(IP_DF);
1073 }
1074 #endif
1075 m->m_pkthdr.len = tlen;
1076 m->m_pkthdr.rcvif = NULL;
1077 #ifdef MAC
1078 if (inp != NULL) {
1079 /*
1080 * Packet is associated with a socket, so allow the
1081 * label of the response to reflect the socket label.
1082 */
1083 INP_WLOCK_ASSERT(inp);
1084 mac_inpcb_create_mbuf(inp, m);
1085 } else {
1086 /*
1087 * Packet is not associated with a socket, so possibly
1088 * update the label in place.
1089 */
1090 mac_netinet_tcp_reply(m);
1091 }
1092 #endif
1093 nth->th_seq = htonl(seq);
1094 nth->th_ack = htonl(ack);
1095 nth->th_x2 = 0;
1096 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1097 nth->th_flags = flags;
1098 if (tp != NULL)
1099 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1100 else
1101 nth->th_win = htons((u_short)win);
1102 nth->th_urp = 0;
1103
1104 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1105 if (to.to_flags & TOF_SIGNATURE) {
1106 if (!TCPMD5_ENABLED() ||
1107 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
1108 m_freem(m);
1109 return;
1110 }
1111 }
1112 #endif
1113
1114 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1115 #ifdef INET6
1116 if (isipv6) {
1117 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1118 nth->th_sum = in6_cksum_pseudo(ip6,
1119 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
1120 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
1121 NULL, NULL);
1122 }
1123 #endif /* INET6 */
1124 #if defined(INET6) && defined(INET)
1125 else
1126 #endif
1127 #ifdef INET
1128 {
1129 m->m_pkthdr.csum_flags = CSUM_TCP;
1130 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1131 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1132 }
1133 #endif /* INET */
1134 #ifdef TCPDEBUG
1135 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1136 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1137 #endif
1138 TCP_PROBE3(debug__output, tp, th, m);
1139 if (flags & TH_RST)
1140 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
1141
1142 #ifdef INET6
1143 if (isipv6) {
1144 TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
1145 (void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
1146 }
1147 #endif /* INET6 */
1148 #if defined(INET) && defined(INET6)
1149 else
1150 #endif
1151 #ifdef INET
1152 {
1153 TCP_PROBE5(send, NULL, tp, ip, tp, nth);
1154 (void)ip_output(m, NULL, NULL, 0, NULL, inp);
1155 }
1156 #endif
1157 }
1158
1159 /*
1160 * Create a new TCP control block, making an
1161 * empty reassembly queue and hooking it to the argument
1162 * protocol control block. The `inp' parameter must have
1163 * come from the zone allocator set up in tcp_init().
1164 */
1165 struct tcpcb *
tcp_newtcpcb(struct inpcb * inp)1166 tcp_newtcpcb(struct inpcb *inp)
1167 {
1168 struct tcpcb_mem *tm;
1169 struct tcpcb *tp;
1170 #ifdef INET6
1171 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1172 #endif /* INET6 */
1173
1174 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1175 if (tm == NULL)
1176 return (NULL);
1177 tp = &tm->tcb;
1178
1179 /* Initialise cc_var struct for this tcpcb. */
1180 tp->ccv = &tm->ccv;
1181 tp->ccv->type = IPPROTO_TCP;
1182 tp->ccv->ccvc.tcp = tp;
1183 rw_rlock(&tcp_function_lock);
1184 tp->t_fb = tcp_func_set_ptr;
1185 refcount_acquire(&tp->t_fb->tfb_refcnt);
1186 rw_runlock(&tcp_function_lock);
1187 if (tp->t_fb->tfb_tcp_fb_init) {
1188 (*tp->t_fb->tfb_tcp_fb_init)(tp);
1189 }
1190 /*
1191 * Use the current system default CC algorithm.
1192 */
1193 CC_LIST_RLOCK();
1194 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1195 CC_ALGO(tp) = CC_DEFAULT();
1196 CC_LIST_RUNLOCK();
1197
1198 if (CC_ALGO(tp)->cb_init != NULL)
1199 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1200 if (tp->t_fb->tfb_tcp_fb_fini)
1201 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1202 refcount_release(&tp->t_fb->tfb_refcnt);
1203 uma_zfree(V_tcpcb_zone, tm);
1204 return (NULL);
1205 }
1206
1207 tp->osd = &tm->osd;
1208 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1209 if (tp->t_fb->tfb_tcp_fb_fini)
1210 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1211 refcount_release(&tp->t_fb->tfb_refcnt);
1212 uma_zfree(V_tcpcb_zone, tm);
1213 return (NULL);
1214 }
1215
1216 #ifdef VIMAGE
1217 tp->t_vnet = inp->inp_vnet;
1218 #endif
1219 tp->t_timers = &tm->tt;
1220 TAILQ_INIT(&tp->t_segq);
1221 tp->t_maxseg =
1222 #ifdef INET6
1223 isipv6 ? V_tcp_v6mssdflt :
1224 #endif /* INET6 */
1225 V_tcp_mssdflt;
1226
1227 /* Set up our timeouts. */
1228 callout_init(&tp->t_timers->tt_rexmt, 1);
1229 callout_init(&tp->t_timers->tt_persist, 1);
1230 callout_init(&tp->t_timers->tt_keep, 1);
1231 callout_init(&tp->t_timers->tt_2msl, 1);
1232 callout_init(&tp->t_timers->tt_delack, 1);
1233
1234 if (V_tcp_do_rfc1323)
1235 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1236 if (V_tcp_do_sack)
1237 tp->t_flags |= TF_SACK_PERMIT;
1238 TAILQ_INIT(&tp->snd_holes);
1239 /*
1240 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1241 * is called.
1242 */
1243 in_pcbref(inp); /* Reference for tcpcb */
1244 tp->t_inpcb = inp;
1245
1246 /*
1247 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1248 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
1249 * reasonable initial retransmit time.
1250 */
1251 tp->t_srtt = TCPTV_SRTTBASE;
1252 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1253 tp->t_rttmin = tcp_rexmit_min;
1254 tp->t_rxtcur = TCPTV_RTOBASE;
1255 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1256 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1257 tp->t_rcvtime = ticks;
1258 /*
1259 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1260 * because the socket may be bound to an IPv6 wildcard address,
1261 * which may match an IPv4-mapped IPv6 address.
1262 */
1263 inp->inp_ip_ttl = V_ip_defttl;
1264 inp->inp_ppcb = tp;
1265 #ifdef TCPPCAP
1266 /*
1267 * Init the TCP PCAP queues.
1268 */
1269 tcp_pcap_tcpcb_init(tp);
1270 #endif
1271 return (tp); /* XXX */
1272 }
1273
1274 /*
1275 * Switch the congestion control algorithm back to NewReno for any active
1276 * control blocks using an algorithm which is about to go away.
1277 * This ensures the CC framework can allow the unload to proceed without leaving
1278 * any dangling pointers which would trigger a panic.
1279 * Returning non-zero would inform the CC framework that something went wrong
1280 * and it would be unsafe to allow the unload to proceed. However, there is no
1281 * way for this to occur with this implementation so we always return zero.
1282 */
1283 int
tcp_ccalgounload(struct cc_algo * unload_algo)1284 tcp_ccalgounload(struct cc_algo *unload_algo)
1285 {
1286 struct cc_algo *tmpalgo;
1287 struct inpcb *inp;
1288 struct tcpcb *tp;
1289 VNET_ITERATOR_DECL(vnet_iter);
1290
1291 /*
1292 * Check all active control blocks across all network stacks and change
1293 * any that are using "unload_algo" back to NewReno. If "unload_algo"
1294 * requires cleanup code to be run, call it.
1295 */
1296 VNET_LIST_RLOCK();
1297 VNET_FOREACH(vnet_iter) {
1298 CURVNET_SET(vnet_iter);
1299 INP_INFO_WLOCK(&V_tcbinfo);
1300 /*
1301 * New connections already part way through being initialised
1302 * with the CC algo we're removing will not race with this code
1303 * because the INP_INFO_WLOCK is held during initialisation. We
1304 * therefore don't enter the loop below until the connection
1305 * list has stabilised.
1306 */
1307 LIST_FOREACH(inp, &V_tcb, inp_list) {
1308 INP_WLOCK(inp);
1309 /* Important to skip tcptw structs. */
1310 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1311 (tp = intotcpcb(inp)) != NULL) {
1312 /*
1313 * By holding INP_WLOCK here, we are assured
1314 * that the connection is not currently
1315 * executing inside the CC module's functions
1316 * i.e. it is safe to make the switch back to
1317 * NewReno.
1318 */
1319 if (CC_ALGO(tp) == unload_algo) {
1320 tmpalgo = CC_ALGO(tp);
1321 /* NewReno does not require any init. */
1322 CC_ALGO(tp) = &newreno_cc_algo;
1323 if (tmpalgo->cb_destroy != NULL)
1324 tmpalgo->cb_destroy(tp->ccv);
1325 }
1326 }
1327 INP_WUNLOCK(inp);
1328 }
1329 INP_INFO_WUNLOCK(&V_tcbinfo);
1330 CURVNET_RESTORE();
1331 }
1332 VNET_LIST_RUNLOCK();
1333
1334 return (0);
1335 }
1336
1337 /*
1338 * Drop a TCP connection, reporting
1339 * the specified error. If connection is synchronized,
1340 * then send a RST to peer.
1341 */
1342 struct tcpcb *
tcp_drop(struct tcpcb * tp,int errno)1343 tcp_drop(struct tcpcb *tp, int errno)
1344 {
1345 struct socket *so = tp->t_inpcb->inp_socket;
1346
1347 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1348 INP_WLOCK_ASSERT(tp->t_inpcb);
1349
1350 if (TCPS_HAVERCVDSYN(tp->t_state)) {
1351 tcp_state_change(tp, TCPS_CLOSED);
1352 (void) tp->t_fb->tfb_tcp_output(tp);
1353 TCPSTAT_INC(tcps_drops);
1354 } else
1355 TCPSTAT_INC(tcps_conndrops);
1356 if (errno == ETIMEDOUT && tp->t_softerror)
1357 errno = tp->t_softerror;
1358 so->so_error = errno;
1359 return (tcp_close(tp));
1360 }
1361
1362 void
tcp_discardcb(struct tcpcb * tp)1363 tcp_discardcb(struct tcpcb *tp)
1364 {
1365 struct inpcb *inp = tp->t_inpcb;
1366 struct socket *so = inp->inp_socket;
1367 #ifdef INET6
1368 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1369 #endif /* INET6 */
1370 int released;
1371
1372 INP_WLOCK_ASSERT(inp);
1373
1374 /*
1375 * Make sure that all of our timers are stopped before we delete the
1376 * PCB.
1377 *
1378 * If stopping a timer fails, we schedule a discard function in same
1379 * callout, and the last discard function called will take care of
1380 * deleting the tcpcb.
1381 */
1382 tp->t_timers->tt_draincnt = 0;
1383 tcp_timer_stop(tp, TT_REXMT);
1384 tcp_timer_stop(tp, TT_PERSIST);
1385 tcp_timer_stop(tp, TT_KEEP);
1386 tcp_timer_stop(tp, TT_2MSL);
1387 tcp_timer_stop(tp, TT_DELACK);
1388 if (tp->t_fb->tfb_tcp_timer_stop_all) {
1389 /*
1390 * Call the stop-all function of the methods,
1391 * this function should call the tcp_timer_stop()
1392 * method with each of the function specific timeouts.
1393 * That stop will be called via the tfb_tcp_timer_stop()
1394 * which should use the async drain function of the
1395 * callout system (see tcp_var.h).
1396 */
1397 tp->t_fb->tfb_tcp_timer_stop_all(tp);
1398 }
1399
1400 /*
1401 * If we got enough samples through the srtt filter,
1402 * save the rtt and rttvar in the routing entry.
1403 * 'Enough' is arbitrarily defined as 4 rtt samples.
1404 * 4 samples is enough for the srtt filter to converge
1405 * to within enough % of the correct value; fewer samples
1406 * and we could save a bogus rtt. The danger is not high
1407 * as tcp quickly recovers from everything.
1408 * XXX: Works very well but needs some more statistics!
1409 */
1410 if (tp->t_rttupdated >= 4) {
1411 struct hc_metrics_lite metrics;
1412 u_long ssthresh;
1413
1414 bzero(&metrics, sizeof(metrics));
1415 /*
1416 * Update the ssthresh always when the conditions below
1417 * are satisfied. This gives us better new start value
1418 * for the congestion avoidance for new connections.
1419 * ssthresh is only set if packet loss occurred on a session.
1420 *
1421 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1422 * being torn down. Ideally this code would not use 'so'.
1423 */
1424 ssthresh = tp->snd_ssthresh;
1425 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1426 /*
1427 * convert the limit from user data bytes to
1428 * packets then to packet data bytes.
1429 */
1430 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1431 if (ssthresh < 2)
1432 ssthresh = 2;
1433 ssthresh *= (u_long)(tp->t_maxseg +
1434 #ifdef INET6
1435 (isipv6 ? sizeof (struct ip6_hdr) +
1436 sizeof (struct tcphdr) :
1437 #endif
1438 sizeof (struct tcpiphdr)
1439 #ifdef INET6
1440 )
1441 #endif
1442 );
1443 } else
1444 ssthresh = 0;
1445 metrics.rmx_ssthresh = ssthresh;
1446
1447 metrics.rmx_rtt = tp->t_srtt;
1448 metrics.rmx_rttvar = tp->t_rttvar;
1449 metrics.rmx_cwnd = tp->snd_cwnd;
1450 metrics.rmx_sendpipe = 0;
1451 metrics.rmx_recvpipe = 0;
1452
1453 tcp_hc_update(&inp->inp_inc, &metrics);
1454 }
1455
1456 /* free the reassembly queue, if any */
1457 tcp_reass_flush(tp);
1458
1459 #ifdef TCP_OFFLOAD
1460 /* Disconnect offload device, if any. */
1461 if (tp->t_flags & TF_TOE)
1462 tcp_offload_detach(tp);
1463 #endif
1464
1465 tcp_free_sackholes(tp);
1466
1467 #ifdef TCPPCAP
1468 /* Free the TCP PCAP queues. */
1469 tcp_pcap_drain(&(tp->t_inpkts));
1470 tcp_pcap_drain(&(tp->t_outpkts));
1471 #endif
1472
1473 /* Allow the CC algorithm to clean up after itself. */
1474 if (CC_ALGO(tp)->cb_destroy != NULL)
1475 CC_ALGO(tp)->cb_destroy(tp->ccv);
1476
1477 khelp_destroy_osd(tp->osd);
1478
1479 CC_ALGO(tp) = NULL;
1480 inp->inp_ppcb = NULL;
1481 if (tp->t_timers->tt_draincnt == 0) {
1482 /* We own the last reference on tcpcb, let's free it. */
1483 TCPSTATES_DEC(tp->t_state);
1484 if (tp->t_fb->tfb_tcp_fb_fini)
1485 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1486 refcount_release(&tp->t_fb->tfb_refcnt);
1487 tp->t_inpcb = NULL;
1488 uma_zfree(V_tcpcb_zone, tp);
1489 released = in_pcbrele_wlocked(inp);
1490 KASSERT(!released, ("%s: inp %p should not have been released "
1491 "here", __func__, inp));
1492 }
1493 }
1494
1495 void
tcp_timer_discard(void * ptp)1496 tcp_timer_discard(void *ptp)
1497 {
1498 struct inpcb *inp;
1499 struct tcpcb *tp;
1500
1501 tp = (struct tcpcb *)ptp;
1502 CURVNET_SET(tp->t_vnet);
1503 INP_INFO_RLOCK(&V_tcbinfo);
1504 inp = tp->t_inpcb;
1505 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1506 __func__, tp));
1507 INP_WLOCK(inp);
1508 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1509 ("%s: tcpcb has to be stopped here", __func__));
1510 tp->t_timers->tt_draincnt--;
1511 if (tp->t_timers->tt_draincnt == 0) {
1512 /* We own the last reference on this tcpcb, let's free it. */
1513 TCPSTATES_DEC(tp->t_state);
1514 if (tp->t_fb->tfb_tcp_fb_fini)
1515 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1516 refcount_release(&tp->t_fb->tfb_refcnt);
1517 tp->t_inpcb = NULL;
1518 uma_zfree(V_tcpcb_zone, tp);
1519 if (in_pcbrele_wlocked(inp)) {
1520 INP_INFO_RUNLOCK(&V_tcbinfo);
1521 CURVNET_RESTORE();
1522 return;
1523 }
1524 }
1525 INP_WUNLOCK(inp);
1526 INP_INFO_RUNLOCK(&V_tcbinfo);
1527 CURVNET_RESTORE();
1528 }
1529
1530 /*
1531 * Attempt to close a TCP control block, marking it as dropped, and freeing
1532 * the socket if we hold the only reference.
1533 */
1534 struct tcpcb *
tcp_close(struct tcpcb * tp)1535 tcp_close(struct tcpcb *tp)
1536 {
1537 struct inpcb *inp = tp->t_inpcb;
1538 struct socket *so;
1539
1540 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1541 INP_WLOCK_ASSERT(inp);
1542
1543 #ifdef TCP_OFFLOAD
1544 if (tp->t_state == TCPS_LISTEN)
1545 tcp_offload_listen_stop(tp);
1546 #endif
1547 #ifdef TCP_RFC7413
1548 /*
1549 * This releases the TFO pending counter resource for TFO listen
1550 * sockets as well as passively-created TFO sockets that transition
1551 * from SYN_RECEIVED to CLOSED.
1552 */
1553 if (tp->t_tfo_pending) {
1554 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
1555 tp->t_tfo_pending = NULL;
1556 }
1557 #endif
1558 in_pcbdrop(inp);
1559 TCPSTAT_INC(tcps_closed);
1560 if (tp->t_state != TCPS_CLOSED)
1561 tcp_state_change(tp, TCPS_CLOSED);
1562 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1563 so = inp->inp_socket;
1564 soisdisconnected(so);
1565 if (inp->inp_flags & INP_SOCKREF) {
1566 KASSERT(so->so_state & SS_PROTOREF,
1567 ("tcp_close: !SS_PROTOREF"));
1568 inp->inp_flags &= ~INP_SOCKREF;
1569 INP_WUNLOCK(inp);
1570 ACCEPT_LOCK();
1571 SOCK_LOCK(so);
1572 so->so_state &= ~SS_PROTOREF;
1573 sofree(so);
1574 return (NULL);
1575 }
1576 return (tp);
1577 }
1578
1579 void
tcp_drain(void)1580 tcp_drain(void)
1581 {
1582 VNET_ITERATOR_DECL(vnet_iter);
1583
1584 if (!do_tcpdrain)
1585 return;
1586
1587 VNET_LIST_RLOCK_NOSLEEP();
1588 VNET_FOREACH(vnet_iter) {
1589 CURVNET_SET(vnet_iter);
1590 struct inpcb *inpb;
1591 struct tcpcb *tcpb;
1592
1593 /*
1594 * Walk the tcpbs, if existing, and flush the reassembly queue,
1595 * if there is one...
1596 * XXX: The "Net/3" implementation doesn't imply that the TCP
1597 * reassembly queue should be flushed, but in a situation
1598 * where we're really low on mbufs, this is potentially
1599 * useful.
1600 */
1601 INP_INFO_WLOCK(&V_tcbinfo);
1602 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1603 if (inpb->inp_flags & INP_TIMEWAIT)
1604 continue;
1605 INP_WLOCK(inpb);
1606 if ((tcpb = intotcpcb(inpb)) != NULL) {
1607 tcp_reass_flush(tcpb);
1608 tcp_clean_sackreport(tcpb);
1609 #ifdef TCPPCAP
1610 if (tcp_pcap_aggressive_free) {
1611 /* Free the TCP PCAP queues. */
1612 tcp_pcap_drain(&(tcpb->t_inpkts));
1613 tcp_pcap_drain(&(tcpb->t_outpkts));
1614 }
1615 #endif
1616 }
1617 INP_WUNLOCK(inpb);
1618 }
1619 INP_INFO_WUNLOCK(&V_tcbinfo);
1620 CURVNET_RESTORE();
1621 }
1622 VNET_LIST_RUNLOCK_NOSLEEP();
1623 }
1624
1625 /*
1626 * Notify a tcp user of an asynchronous error;
1627 * store error as soft error, but wake up user
1628 * (for now, won't do anything until can select for soft error).
1629 *
1630 * Do not wake up user since there currently is no mechanism for
1631 * reporting soft errors (yet - a kqueue filter may be added).
1632 */
1633 static struct inpcb *
tcp_notify(struct inpcb * inp,int error)1634 tcp_notify(struct inpcb *inp, int error)
1635 {
1636 struct tcpcb *tp;
1637
1638 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1639 INP_WLOCK_ASSERT(inp);
1640
1641 if ((inp->inp_flags & INP_TIMEWAIT) ||
1642 (inp->inp_flags & INP_DROPPED))
1643 return (inp);
1644
1645 tp = intotcpcb(inp);
1646 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1647
1648 /*
1649 * Ignore some errors if we are hooked up.
1650 * If connection hasn't completed, has retransmitted several times,
1651 * and receives a second error, give up now. This is better
1652 * than waiting a long time to establish a connection that
1653 * can never complete.
1654 */
1655 if (tp->t_state == TCPS_ESTABLISHED &&
1656 (error == EHOSTUNREACH || error == ENETUNREACH ||
1657 error == EHOSTDOWN)) {
1658 if (inp->inp_route.ro_rt) {
1659 RTFREE(inp->inp_route.ro_rt);
1660 inp->inp_route.ro_rt = (struct rtentry *)NULL;
1661 }
1662 return (inp);
1663 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1664 tp->t_softerror) {
1665 tp = tcp_drop(tp, error);
1666 if (tp != NULL)
1667 return (inp);
1668 else
1669 return (NULL);
1670 } else {
1671 tp->t_softerror = error;
1672 return (inp);
1673 }
1674 #if 0
1675 wakeup( &so->so_timeo);
1676 sorwakeup(so);
1677 sowwakeup(so);
1678 #endif
1679 }
1680
1681 static int
tcp_pcblist(SYSCTL_HANDLER_ARGS)1682 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1683 {
1684 int error, i, m, n, pcb_count;
1685 struct inpcb *inp, **inp_list;
1686 inp_gen_t gencnt;
1687 struct xinpgen xig;
1688
1689 /*
1690 * The process of preparing the TCB list is too time-consuming and
1691 * resource-intensive to repeat twice on every request.
1692 */
1693 if (req->oldptr == NULL) {
1694 n = V_tcbinfo.ipi_count +
1695 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
1696 n += imax(n / 8, 10);
1697 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1698 return (0);
1699 }
1700
1701 if (req->newptr != NULL)
1702 return (EPERM);
1703
1704 /*
1705 * OK, now we're committed to doing something.
1706 */
1707 INP_LIST_RLOCK(&V_tcbinfo);
1708 gencnt = V_tcbinfo.ipi_gencnt;
1709 n = V_tcbinfo.ipi_count;
1710 INP_LIST_RUNLOCK(&V_tcbinfo);
1711
1712 m = counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
1713
1714 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1715 + (n + m) * sizeof(struct xtcpcb));
1716 if (error != 0)
1717 return (error);
1718
1719 bzero(&xig, sizeof(xig));
1720 xig.xig_len = sizeof xig;
1721 xig.xig_count = n + m;
1722 xig.xig_gen = gencnt;
1723 xig.xig_sogen = so_gencnt;
1724 error = SYSCTL_OUT(req, &xig, sizeof xig);
1725 if (error)
1726 return (error);
1727
1728 error = syncache_pcblist(req, m, &pcb_count);
1729 if (error)
1730 return (error);
1731
1732 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1733
1734 INP_INFO_WLOCK(&V_tcbinfo);
1735 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1736 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1737 INP_WLOCK(inp);
1738 if (inp->inp_gencnt <= gencnt) {
1739 /*
1740 * XXX: This use of cr_cansee(), introduced with
1741 * TCP state changes, is not quite right, but for
1742 * now, better than nothing.
1743 */
1744 if (inp->inp_flags & INP_TIMEWAIT) {
1745 if (intotw(inp) != NULL)
1746 error = cr_cansee(req->td->td_ucred,
1747 intotw(inp)->tw_cred);
1748 else
1749 error = EINVAL; /* Skip this inp. */
1750 } else
1751 error = cr_canseeinpcb(req->td->td_ucred, inp);
1752 if (error == 0) {
1753 in_pcbref(inp);
1754 inp_list[i++] = inp;
1755 }
1756 }
1757 INP_WUNLOCK(inp);
1758 }
1759 INP_INFO_WUNLOCK(&V_tcbinfo);
1760 n = i;
1761
1762 error = 0;
1763 for (i = 0; i < n; i++) {
1764 inp = inp_list[i];
1765 INP_RLOCK(inp);
1766 if (inp->inp_gencnt <= gencnt) {
1767 struct xtcpcb xt;
1768 void *inp_ppcb;
1769
1770 bzero(&xt, sizeof(xt));
1771 xt.xt_len = sizeof xt;
1772 /* XXX should avoid extra copy */
1773 bcopy(inp, &xt.xt_inp, sizeof *inp);
1774 inp_ppcb = inp->inp_ppcb;
1775 if (inp_ppcb == NULL)
1776 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1777 else if (inp->inp_flags & INP_TIMEWAIT) {
1778 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1779 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1780 } else {
1781 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1782 if (xt.xt_tp.t_timers)
1783 tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1784 }
1785 if (inp->inp_socket != NULL)
1786 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1787 else {
1788 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1789 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1790 }
1791 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1792 INP_RUNLOCK(inp);
1793 error = SYSCTL_OUT(req, &xt, sizeof xt);
1794 } else
1795 INP_RUNLOCK(inp);
1796 }
1797 INP_INFO_RLOCK(&V_tcbinfo);
1798 for (i = 0; i < n; i++) {
1799 inp = inp_list[i];
1800 INP_RLOCK(inp);
1801 if (!in_pcbrele_rlocked(inp))
1802 INP_RUNLOCK(inp);
1803 }
1804 INP_INFO_RUNLOCK(&V_tcbinfo);
1805
1806 if (!error) {
1807 /*
1808 * Give the user an updated idea of our state.
1809 * If the generation differs from what we told
1810 * her before, she knows that something happened
1811 * while we were processing this request, and it
1812 * might be necessary to retry.
1813 */
1814 INP_LIST_RLOCK(&V_tcbinfo);
1815 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1816 xig.xig_sogen = so_gencnt;
1817 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1818 INP_LIST_RUNLOCK(&V_tcbinfo);
1819 error = SYSCTL_OUT(req, &xig, sizeof xig);
1820 }
1821 free(inp_list, M_TEMP);
1822 return (error);
1823 }
1824
1825 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1826 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1827 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1828
1829 #ifdef INET
1830 static int
tcp_getcred(SYSCTL_HANDLER_ARGS)1831 tcp_getcred(SYSCTL_HANDLER_ARGS)
1832 {
1833 struct xucred xuc;
1834 struct sockaddr_in addrs[2];
1835 struct inpcb *inp;
1836 int error;
1837
1838 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1839 if (error)
1840 return (error);
1841 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1842 if (error)
1843 return (error);
1844 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1845 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1846 if (inp != NULL) {
1847 if (inp->inp_socket == NULL)
1848 error = ENOENT;
1849 if (error == 0)
1850 error = cr_canseeinpcb(req->td->td_ucred, inp);
1851 if (error == 0)
1852 cru2x(inp->inp_cred, &xuc);
1853 INP_RUNLOCK(inp);
1854 } else
1855 error = ENOENT;
1856 if (error == 0)
1857 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1858 return (error);
1859 }
1860
1861 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1862 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1863 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1864 #endif /* INET */
1865
1866 #ifdef INET6
1867 static int
tcp6_getcred(SYSCTL_HANDLER_ARGS)1868 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1869 {
1870 struct xucred xuc;
1871 struct sockaddr_in6 addrs[2];
1872 struct inpcb *inp;
1873 int error;
1874 #ifdef INET
1875 int mapped = 0;
1876 #endif
1877
1878 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1879 if (error)
1880 return (error);
1881 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1882 if (error)
1883 return (error);
1884 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1885 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1886 return (error);
1887 }
1888 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1889 #ifdef INET
1890 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1891 mapped = 1;
1892 else
1893 #endif
1894 return (EINVAL);
1895 }
1896
1897 #ifdef INET
1898 if (mapped == 1)
1899 inp = in_pcblookup(&V_tcbinfo,
1900 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1901 addrs[1].sin6_port,
1902 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1903 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1904 else
1905 #endif
1906 inp = in6_pcblookup(&V_tcbinfo,
1907 &addrs[1].sin6_addr, addrs[1].sin6_port,
1908 &addrs[0].sin6_addr, addrs[0].sin6_port,
1909 INPLOOKUP_RLOCKPCB, NULL);
1910 if (inp != NULL) {
1911 if (inp->inp_socket == NULL)
1912 error = ENOENT;
1913 if (error == 0)
1914 error = cr_canseeinpcb(req->td->td_ucred, inp);
1915 if (error == 0)
1916 cru2x(inp->inp_cred, &xuc);
1917 INP_RUNLOCK(inp);
1918 } else
1919 error = ENOENT;
1920 if (error == 0)
1921 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1922 return (error);
1923 }
1924
1925 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1926 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1927 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1928 #endif /* INET6 */
1929
1930
1931 #ifdef INET
1932 void
tcp_ctlinput(int cmd,struct sockaddr * sa,void * vip)1933 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1934 {
1935 struct ip *ip = vip;
1936 struct tcphdr *th;
1937 struct in_addr faddr;
1938 struct inpcb *inp;
1939 struct tcpcb *tp;
1940 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1941 struct icmp *icp;
1942 struct in_conninfo inc;
1943 tcp_seq icmp_tcp_seq;
1944 int mtu;
1945
1946 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1947 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1948 return;
1949
1950 if (cmd == PRC_MSGSIZE)
1951 notify = tcp_mtudisc_notify;
1952 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1953 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
1954 cmd == PRC_TIMXCEED_INTRANS) && ip)
1955 notify = tcp_drop_syn_sent;
1956
1957 /*
1958 * Hostdead is ugly because it goes linearly through all PCBs.
1959 * XXX: We never get this from ICMP, otherwise it makes an
1960 * excellent DoS attack on machines with many connections.
1961 */
1962 else if (cmd == PRC_HOSTDEAD)
1963 ip = NULL;
1964 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1965 return;
1966
1967 if (ip == NULL) {
1968 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1969 return;
1970 }
1971
1972 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
1973 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1974 INP_INFO_RLOCK(&V_tcbinfo);
1975 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
1976 th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1977 if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
1978 /* signal EHOSTDOWN, as it flushes the cached route */
1979 inp = (*notify)(inp, EHOSTDOWN);
1980 goto out;
1981 }
1982 icmp_tcp_seq = th->th_seq;
1983 if (inp != NULL) {
1984 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1985 !(inp->inp_flags & INP_DROPPED) &&
1986 !(inp->inp_socket == NULL)) {
1987 tp = intotcpcb(inp);
1988 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
1989 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
1990 if (cmd == PRC_MSGSIZE) {
1991 /*
1992 * MTU discovery:
1993 * If we got a needfrag set the MTU
1994 * in the route to the suggested new
1995 * value (if given) and then notify.
1996 */
1997 mtu = ntohs(icp->icmp_nextmtu);
1998 /*
1999 * If no alternative MTU was
2000 * proposed, try the next smaller
2001 * one.
2002 */
2003 if (!mtu)
2004 mtu = ip_next_mtu(
2005 ntohs(ip->ip_len), 1);
2006 if (mtu < V_tcp_minmss +
2007 sizeof(struct tcpiphdr))
2008 mtu = V_tcp_minmss +
2009 sizeof(struct tcpiphdr);
2010 /*
2011 * Only process the offered MTU if it
2012 * is smaller than the current one.
2013 */
2014 if (mtu < tp->t_maxseg +
2015 sizeof(struct tcpiphdr)) {
2016 bzero(&inc, sizeof(inc));
2017 inc.inc_faddr = faddr;
2018 inc.inc_fibnum =
2019 inp->inp_inc.inc_fibnum;
2020 tcp_hc_updatemtu(&inc, mtu);
2021 tcp_mtudisc(inp, mtu);
2022 }
2023 } else
2024 inp = (*notify)(inp,
2025 inetctlerrmap[cmd]);
2026 }
2027 }
2028 } else {
2029 bzero(&inc, sizeof(inc));
2030 inc.inc_fport = th->th_dport;
2031 inc.inc_lport = th->th_sport;
2032 inc.inc_faddr = faddr;
2033 inc.inc_laddr = ip->ip_src;
2034 syncache_unreach(&inc, icmp_tcp_seq);
2035 }
2036 out:
2037 if (inp != NULL)
2038 INP_WUNLOCK(inp);
2039 INP_INFO_RUNLOCK(&V_tcbinfo);
2040 }
2041 #endif /* INET */
2042
2043 #ifdef INET6
2044 void
tcp6_ctlinput(int cmd,struct sockaddr * sa,void * d)2045 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
2046 {
2047 struct in6_addr *dst;
2048 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2049 struct ip6_hdr *ip6;
2050 struct mbuf *m;
2051 struct inpcb *inp;
2052 struct tcpcb *tp;
2053 struct icmp6_hdr *icmp6;
2054 struct ip6ctlparam *ip6cp = NULL;
2055 const struct sockaddr_in6 *sa6_src = NULL;
2056 struct in_conninfo inc;
2057 struct tcp_ports {
2058 uint16_t th_sport;
2059 uint16_t th_dport;
2060 } t_ports;
2061 tcp_seq icmp_tcp_seq;
2062 unsigned int mtu;
2063 unsigned int off;
2064
2065 if (sa->sa_family != AF_INET6 ||
2066 sa->sa_len != sizeof(struct sockaddr_in6))
2067 return;
2068
2069 /* if the parameter is from icmp6, decode it. */
2070 if (d != NULL) {
2071 ip6cp = (struct ip6ctlparam *)d;
2072 icmp6 = ip6cp->ip6c_icmp6;
2073 m = ip6cp->ip6c_m;
2074 ip6 = ip6cp->ip6c_ip6;
2075 off = ip6cp->ip6c_off;
2076 sa6_src = ip6cp->ip6c_src;
2077 dst = ip6cp->ip6c_finaldst;
2078 } else {
2079 m = NULL;
2080 ip6 = NULL;
2081 off = 0; /* fool gcc */
2082 sa6_src = &sa6_any;
2083 dst = NULL;
2084 }
2085
2086 if (cmd == PRC_MSGSIZE)
2087 notify = tcp_mtudisc_notify;
2088 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2089 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2090 cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
2091 notify = tcp_drop_syn_sent;
2092
2093 /*
2094 * Hostdead is ugly because it goes linearly through all PCBs.
2095 * XXX: We never get this from ICMP, otherwise it makes an
2096 * excellent DoS attack on machines with many connections.
2097 */
2098 else if (cmd == PRC_HOSTDEAD)
2099 ip6 = NULL;
2100 else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
2101 return;
2102
2103 if (ip6 == NULL) {
2104 in6_pcbnotify(&V_tcbinfo, sa, 0,
2105 (const struct sockaddr *)sa6_src,
2106 0, cmd, NULL, notify);
2107 return;
2108 }
2109
2110 /* Check if we can safely get the ports from the tcp hdr */
2111 if (m == NULL ||
2112 (m->m_pkthdr.len <
2113 (int32_t) (off + sizeof(struct tcp_ports)))) {
2114 return;
2115 }
2116 bzero(&t_ports, sizeof(struct tcp_ports));
2117 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
2118 INP_INFO_RLOCK(&V_tcbinfo);
2119 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
2120 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
2121 if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2122 /* signal EHOSTDOWN, as it flushes the cached route */
2123 inp = (*notify)(inp, EHOSTDOWN);
2124 goto out;
2125 }
2126 off += sizeof(struct tcp_ports);
2127 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
2128 goto out;
2129 }
2130 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
2131 if (inp != NULL) {
2132 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2133 !(inp->inp_flags & INP_DROPPED) &&
2134 !(inp->inp_socket == NULL)) {
2135 tp = intotcpcb(inp);
2136 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2137 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2138 if (cmd == PRC_MSGSIZE) {
2139 /*
2140 * MTU discovery:
2141 * If we got a needfrag set the MTU
2142 * in the route to the suggested new
2143 * value (if given) and then notify.
2144 */
2145 mtu = ntohl(icmp6->icmp6_mtu);
2146 /*
2147 * If no alternative MTU was
2148 * proposed, or the proposed
2149 * MTU was too small, set to
2150 * the min.
2151 */
2152 if (mtu < IPV6_MMTU)
2153 mtu = IPV6_MMTU - 8;
2154 bzero(&inc, sizeof(inc));
2155 inc.inc_fibnum = M_GETFIB(m);
2156 inc.inc_flags |= INC_ISIPV6;
2157 inc.inc6_faddr = *dst;
2158 if (in6_setscope(&inc.inc6_faddr,
2159 m->m_pkthdr.rcvif, NULL))
2160 goto out;
2161 /*
2162 * Only process the offered MTU if it
2163 * is smaller than the current one.
2164 */
2165 if (mtu < tp->t_maxseg +
2166 sizeof (struct tcphdr) +
2167 sizeof (struct ip6_hdr)) {
2168 tcp_hc_updatemtu(&inc, mtu);
2169 tcp_mtudisc(inp, mtu);
2170 ICMP6STAT_INC(icp6s_pmtuchg);
2171 }
2172 } else
2173 inp = (*notify)(inp,
2174 inet6ctlerrmap[cmd]);
2175 }
2176 }
2177 } else {
2178 bzero(&inc, sizeof(inc));
2179 inc.inc_fibnum = M_GETFIB(m);
2180 inc.inc_flags |= INC_ISIPV6;
2181 inc.inc_fport = t_ports.th_dport;
2182 inc.inc_lport = t_ports.th_sport;
2183 inc.inc6_faddr = *dst;
2184 inc.inc6_laddr = ip6->ip6_src;
2185 syncache_unreach(&inc, icmp_tcp_seq);
2186 }
2187 out:
2188 if (inp != NULL)
2189 INP_WUNLOCK(inp);
2190 INP_INFO_RUNLOCK(&V_tcbinfo);
2191 }
2192 #endif /* INET6 */
2193
2194 static uint32_t
tcp_keyed_hash(struct in_conninfo * inc,u_char * key)2195 tcp_keyed_hash(struct in_conninfo *inc, u_char *key)
2196 {
2197 MD5_CTX ctx;
2198 uint32_t hash[4];
2199
2200 MD5Init(&ctx);
2201 MD5Update(&ctx, &inc->inc_fport, sizeof(uint16_t));
2202 MD5Update(&ctx, &inc->inc_lport, sizeof(uint16_t));
2203 switch (inc->inc_flags & INC_ISIPV6) {
2204 #ifdef INET
2205 case 0:
2206 MD5Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr));
2207 MD5Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr));
2208 break;
2209 #endif
2210 #ifdef INET6
2211 case INC_ISIPV6:
2212 MD5Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr));
2213 MD5Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr));
2214 break;
2215 #endif
2216 }
2217 MD5Update(&ctx, key, 32);
2218 MD5Final((unsigned char *)hash, &ctx);
2219
2220 return (hash[0]);
2221 }
2222
2223 uint32_t
tcp_new_ts_offset(struct in_conninfo * inc)2224 tcp_new_ts_offset(struct in_conninfo *inc)
2225 {
2226 return (tcp_keyed_hash(inc, V_ts_offset_secret));
2227 }
2228
2229 /*
2230 * Following is where TCP initial sequence number generation occurs.
2231 *
2232 * There are two places where we must use initial sequence numbers:
2233 * 1. In SYN-ACK packets.
2234 * 2. In SYN packets.
2235 *
2236 * All ISNs for SYN-ACK packets are generated by the syncache. See
2237 * tcp_syncache.c for details.
2238 *
2239 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2240 * depends on this property. In addition, these ISNs should be
2241 * unguessable so as to prevent connection hijacking. To satisfy
2242 * the requirements of this situation, the algorithm outlined in
2243 * RFC 1948 is used, with only small modifications.
2244 *
2245 * Implementation details:
2246 *
2247 * Time is based off the system timer, and is corrected so that it
2248 * increases by one megabyte per second. This allows for proper
2249 * recycling on high speed LANs while still leaving over an hour
2250 * before rollover.
2251 *
2252 * As reading the *exact* system time is too expensive to be done
2253 * whenever setting up a TCP connection, we increment the time
2254 * offset in two ways. First, a small random positive increment
2255 * is added to isn_offset for each connection that is set up.
2256 * Second, the function tcp_isn_tick fires once per clock tick
2257 * and increments isn_offset as necessary so that sequence numbers
2258 * are incremented at approximately ISN_BYTES_PER_SECOND. The
2259 * random positive increments serve only to ensure that the same
2260 * exact sequence number is never sent out twice (as could otherwise
2261 * happen when a port is recycled in less than the system tick
2262 * interval.)
2263 *
2264 * net.inet.tcp.isn_reseed_interval controls the number of seconds
2265 * between seeding of isn_secret. This is normally set to zero,
2266 * as reseeding should not be necessary.
2267 *
2268 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2269 * isn_offset_old, and isn_ctx is performed using the ISN lock. In
2270 * general, this means holding an exclusive (write) lock.
2271 */
2272
2273 #define ISN_BYTES_PER_SECOND 1048576
2274 #define ISN_STATIC_INCREMENT 4096
2275 #define ISN_RANDOM_INCREMENT (4096 - 1)
2276
2277 static VNET_DEFINE(u_char, isn_secret[32]);
2278 static VNET_DEFINE(int, isn_last);
2279 static VNET_DEFINE(int, isn_last_reseed);
2280 static VNET_DEFINE(u_int32_t, isn_offset);
2281 static VNET_DEFINE(u_int32_t, isn_offset_old);
2282
2283 #define V_isn_secret VNET(isn_secret)
2284 #define V_isn_last VNET(isn_last)
2285 #define V_isn_last_reseed VNET(isn_last_reseed)
2286 #define V_isn_offset VNET(isn_offset)
2287 #define V_isn_offset_old VNET(isn_offset_old)
2288
2289 tcp_seq
tcp_new_isn(struct in_conninfo * inc)2290 tcp_new_isn(struct in_conninfo *inc)
2291 {
2292 tcp_seq new_isn;
2293 u_int32_t projected_offset;
2294
2295 ISN_LOCK();
2296 /* Seed if this is the first use, reseed if requested. */
2297 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2298 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2299 < (u_int)ticks))) {
2300 read_random(&V_isn_secret, sizeof(V_isn_secret));
2301 V_isn_last_reseed = ticks;
2302 }
2303
2304 /* Compute the md5 hash and return the ISN. */
2305 new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret);
2306 V_isn_offset += ISN_STATIC_INCREMENT +
2307 (arc4random() & ISN_RANDOM_INCREMENT);
2308 if (ticks != V_isn_last) {
2309 projected_offset = V_isn_offset_old +
2310 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2311 if (SEQ_GT(projected_offset, V_isn_offset))
2312 V_isn_offset = projected_offset;
2313 V_isn_offset_old = V_isn_offset;
2314 V_isn_last = ticks;
2315 }
2316 new_isn += V_isn_offset;
2317 ISN_UNLOCK();
2318 return (new_isn);
2319 }
2320
2321 /*
2322 * When a specific ICMP unreachable message is received and the
2323 * connection state is SYN-SENT, drop the connection. This behavior
2324 * is controlled by the icmp_may_rst sysctl.
2325 */
2326 struct inpcb *
tcp_drop_syn_sent(struct inpcb * inp,int errno)2327 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2328 {
2329 struct tcpcb *tp;
2330
2331 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2332 INP_WLOCK_ASSERT(inp);
2333
2334 if ((inp->inp_flags & INP_TIMEWAIT) ||
2335 (inp->inp_flags & INP_DROPPED))
2336 return (inp);
2337
2338 tp = intotcpcb(inp);
2339 if (tp->t_state != TCPS_SYN_SENT)
2340 return (inp);
2341
2342 tp = tcp_drop(tp, errno);
2343 if (tp != NULL)
2344 return (inp);
2345 else
2346 return (NULL);
2347 }
2348
2349 /*
2350 * When `need fragmentation' ICMP is received, update our idea of the MSS
2351 * based on the new value. Also nudge TCP to send something, since we
2352 * know the packet we just sent was dropped.
2353 * This duplicates some code in the tcp_mss() function in tcp_input.c.
2354 */
2355 static struct inpcb *
tcp_mtudisc_notify(struct inpcb * inp,int error)2356 tcp_mtudisc_notify(struct inpcb *inp, int error)
2357 {
2358
2359 tcp_mtudisc(inp, -1);
2360 return (inp);
2361 }
2362
2363 static void
tcp_mtudisc(struct inpcb * inp,int mtuoffer)2364 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2365 {
2366 struct tcpcb *tp;
2367 struct socket *so;
2368
2369 INP_WLOCK_ASSERT(inp);
2370 if ((inp->inp_flags & INP_TIMEWAIT) ||
2371 (inp->inp_flags & INP_DROPPED))
2372 return;
2373
2374 tp = intotcpcb(inp);
2375 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2376
2377 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2378
2379 so = inp->inp_socket;
2380 SOCKBUF_LOCK(&so->so_snd);
2381 /* If the mss is larger than the socket buffer, decrease the mss. */
2382 if (so->so_snd.sb_hiwat < tp->t_maxseg)
2383 tp->t_maxseg = so->so_snd.sb_hiwat;
2384 SOCKBUF_UNLOCK(&so->so_snd);
2385
2386 TCPSTAT_INC(tcps_mturesent);
2387 tp->t_rtttime = 0;
2388 tp->snd_nxt = tp->snd_una;
2389 tcp_free_sackholes(tp);
2390 tp->snd_recover = tp->snd_max;
2391 if (tp->t_flags & TF_SACK_PERMIT)
2392 EXIT_FASTRECOVERY(tp->t_flags);
2393 tp->t_fb->tfb_tcp_output(tp);
2394 }
2395
2396 #ifdef INET
2397 /*
2398 * Look-up the routing entry to the peer of this inpcb. If no route
2399 * is found and it cannot be allocated, then return 0. This routine
2400 * is called by TCP routines that access the rmx structure and by
2401 * tcp_mss_update to get the peer/interface MTU.
2402 */
2403 u_long
tcp_maxmtu(struct in_conninfo * inc,struct tcp_ifcap * cap)2404 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2405 {
2406 struct nhop4_extended nh4;
2407 struct ifnet *ifp;
2408 u_long maxmtu = 0;
2409
2410 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2411
2412 if (inc->inc_faddr.s_addr != INADDR_ANY) {
2413
2414 if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr,
2415 NHR_REF, 0, &nh4) != 0)
2416 return (0);
2417
2418 ifp = nh4.nh_ifp;
2419 maxmtu = nh4.nh_mtu;
2420
2421 /* Report additional interface capabilities. */
2422 if (cap != NULL) {
2423 if (ifp->if_capenable & IFCAP_TSO4 &&
2424 ifp->if_hwassist & CSUM_TSO) {
2425 cap->ifcap |= CSUM_TSO;
2426 cap->tsomax = ifp->if_hw_tsomax;
2427 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2428 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2429 }
2430 }
2431 fib4_free_nh_ext(inc->inc_fibnum, &nh4);
2432 }
2433 return (maxmtu);
2434 }
2435 #endif /* INET */
2436
2437 #ifdef INET6
2438 u_long
tcp_maxmtu6(struct in_conninfo * inc,struct tcp_ifcap * cap)2439 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2440 {
2441 struct nhop6_extended nh6;
2442 struct in6_addr dst6;
2443 uint32_t scopeid;
2444 struct ifnet *ifp;
2445 u_long maxmtu = 0;
2446
2447 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2448
2449 if (inc->inc_flags & INC_IPV6MINMTU)
2450 return (IPV6_MMTU);
2451
2452 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2453 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
2454 if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0,
2455 0, &nh6) != 0)
2456 return (0);
2457
2458 ifp = nh6.nh_ifp;
2459 maxmtu = nh6.nh_mtu;
2460
2461 /* Report additional interface capabilities. */
2462 if (cap != NULL) {
2463 if (ifp->if_capenable & IFCAP_TSO6 &&
2464 ifp->if_hwassist & CSUM_TSO) {
2465 cap->ifcap |= CSUM_TSO;
2466 cap->tsomax = ifp->if_hw_tsomax;
2467 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2468 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2469 }
2470 }
2471 fib6_free_nh_ext(inc->inc_fibnum, &nh6);
2472 }
2473
2474 return (maxmtu);
2475 }
2476 #endif /* INET6 */
2477
2478 /*
2479 * Calculate effective SMSS per RFC5681 definition for a given TCP
2480 * connection at its current state, taking into account SACK and etc.
2481 */
2482 u_int
tcp_maxseg(const struct tcpcb * tp)2483 tcp_maxseg(const struct tcpcb *tp)
2484 {
2485 u_int optlen;
2486
2487 if (tp->t_flags & TF_NOOPT)
2488 return (tp->t_maxseg);
2489
2490 /*
2491 * Here we have a simplified code from tcp_addoptions(),
2492 * without a proper loop, and having most of paddings hardcoded.
2493 * We might make mistakes with padding here in some edge cases,
2494 * but this is harmless, since result of tcp_maxseg() is used
2495 * only in cwnd and ssthresh estimations.
2496 */
2497 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4)
2498 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2499 if (tp->t_flags & TF_RCVD_TSTMP)
2500 optlen = TCPOLEN_TSTAMP_APPA;
2501 else
2502 optlen = 0;
2503 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2504 if (tp->t_flags & TF_SIGNATURE)
2505 optlen += PAD(TCPOLEN_SIGNATURE);
2506 #endif
2507 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
2508 optlen += TCPOLEN_SACKHDR;
2509 optlen += tp->rcv_numsacks * TCPOLEN_SACK;
2510 optlen = PAD(optlen);
2511 }
2512 } else {
2513 if (tp->t_flags & TF_REQ_TSTMP)
2514 optlen = TCPOLEN_TSTAMP_APPA;
2515 else
2516 optlen = PAD(TCPOLEN_MAXSEG);
2517 if (tp->t_flags & TF_REQ_SCALE)
2518 optlen += PAD(TCPOLEN_WINDOW);
2519 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2520 if (tp->t_flags & TF_SIGNATURE)
2521 optlen += PAD(TCPOLEN_SIGNATURE);
2522 #endif
2523 if (tp->t_flags & TF_SACK_PERMIT)
2524 optlen += PAD(TCPOLEN_SACK_PERMITTED);
2525 }
2526 #undef PAD
2527 optlen = min(optlen, TCP_MAXOLEN);
2528 return (tp->t_maxseg - optlen);
2529 }
2530
2531 static int
sysctl_drop(SYSCTL_HANDLER_ARGS)2532 sysctl_drop(SYSCTL_HANDLER_ARGS)
2533 {
2534 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2535 struct sockaddr_storage addrs[2];
2536 struct inpcb *inp;
2537 struct tcpcb *tp;
2538 struct tcptw *tw;
2539 struct sockaddr_in *fin, *lin;
2540 #ifdef INET6
2541 struct sockaddr_in6 *fin6, *lin6;
2542 #endif
2543 int error;
2544
2545 inp = NULL;
2546 fin = lin = NULL;
2547 #ifdef INET6
2548 fin6 = lin6 = NULL;
2549 #endif
2550 error = 0;
2551
2552 if (req->oldptr != NULL || req->oldlen != 0)
2553 return (EINVAL);
2554 if (req->newptr == NULL)
2555 return (EPERM);
2556 if (req->newlen < sizeof(addrs))
2557 return (ENOMEM);
2558 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2559 if (error)
2560 return (error);
2561
2562 switch (addrs[0].ss_family) {
2563 #ifdef INET6
2564 case AF_INET6:
2565 fin6 = (struct sockaddr_in6 *)&addrs[0];
2566 lin6 = (struct sockaddr_in6 *)&addrs[1];
2567 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2568 lin6->sin6_len != sizeof(struct sockaddr_in6))
2569 return (EINVAL);
2570 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2571 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2572 return (EINVAL);
2573 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2574 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2575 fin = (struct sockaddr_in *)&addrs[0];
2576 lin = (struct sockaddr_in *)&addrs[1];
2577 break;
2578 }
2579 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2580 if (error)
2581 return (error);
2582 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2583 if (error)
2584 return (error);
2585 break;
2586 #endif
2587 #ifdef INET
2588 case AF_INET:
2589 fin = (struct sockaddr_in *)&addrs[0];
2590 lin = (struct sockaddr_in *)&addrs[1];
2591 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2592 lin->sin_len != sizeof(struct sockaddr_in))
2593 return (EINVAL);
2594 break;
2595 #endif
2596 default:
2597 return (EINVAL);
2598 }
2599 INP_INFO_RLOCK(&V_tcbinfo);
2600 switch (addrs[0].ss_family) {
2601 #ifdef INET6
2602 case AF_INET6:
2603 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2604 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2605 INPLOOKUP_WLOCKPCB, NULL);
2606 break;
2607 #endif
2608 #ifdef INET
2609 case AF_INET:
2610 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2611 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2612 break;
2613 #endif
2614 }
2615 if (inp != NULL) {
2616 if (inp->inp_flags & INP_TIMEWAIT) {
2617 /*
2618 * XXXRW: There currently exists a state where an
2619 * inpcb is present, but its timewait state has been
2620 * discarded. For now, don't allow dropping of this
2621 * type of inpcb.
2622 */
2623 tw = intotw(inp);
2624 if (tw != NULL)
2625 tcp_twclose(tw, 0);
2626 else
2627 INP_WUNLOCK(inp);
2628 } else if (!(inp->inp_flags & INP_DROPPED) &&
2629 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2630 tp = intotcpcb(inp);
2631 tp = tcp_drop(tp, ECONNABORTED);
2632 if (tp != NULL)
2633 INP_WUNLOCK(inp);
2634 } else
2635 INP_WUNLOCK(inp);
2636 } else
2637 error = ESRCH;
2638 INP_INFO_RUNLOCK(&V_tcbinfo);
2639 return (error);
2640 }
2641
2642 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2643 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
2644 0, sysctl_drop, "", "Drop TCP connection");
2645
2646 /*
2647 * Generate a standardized TCP log line for use throughout the
2648 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2649 * allow use in the interrupt context.
2650 *
2651 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2652 * NB: The function may return NULL if memory allocation failed.
2653 *
2654 * Due to header inclusion and ordering limitations the struct ip
2655 * and ip6_hdr pointers have to be passed as void pointers.
2656 */
2657 char *
tcp_log_vain(struct in_conninfo * inc,struct tcphdr * th,void * ip4hdr,const void * ip6hdr)2658 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2659 const void *ip6hdr)
2660 {
2661
2662 /* Is logging enabled? */
2663 if (tcp_log_in_vain == 0)
2664 return (NULL);
2665
2666 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2667 }
2668
2669 char *
tcp_log_addrs(struct in_conninfo * inc,struct tcphdr * th,void * ip4hdr,const void * ip6hdr)2670 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2671 const void *ip6hdr)
2672 {
2673
2674 /* Is logging enabled? */
2675 if (tcp_log_debug == 0)
2676 return (NULL);
2677
2678 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2679 }
2680
2681 static char *
tcp_log_addr(struct in_conninfo * inc,struct tcphdr * th,void * ip4hdr,const void * ip6hdr)2682 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2683 const void *ip6hdr)
2684 {
2685 char *s, *sp;
2686 size_t size;
2687 struct ip *ip;
2688 #ifdef INET6
2689 const struct ip6_hdr *ip6;
2690
2691 ip6 = (const struct ip6_hdr *)ip6hdr;
2692 #endif /* INET6 */
2693 ip = (struct ip *)ip4hdr;
2694
2695 /*
2696 * The log line looks like this:
2697 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2698 */
2699 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2700 sizeof(PRINT_TH_FLAGS) + 1 +
2701 #ifdef INET6
2702 2 * INET6_ADDRSTRLEN;
2703 #else
2704 2 * INET_ADDRSTRLEN;
2705 #endif /* INET6 */
2706
2707 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2708 if (s == NULL)
2709 return (NULL);
2710
2711 strcat(s, "TCP: [");
2712 sp = s + strlen(s);
2713
2714 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2715 inet_ntoa_r(inc->inc_faddr, sp);
2716 sp = s + strlen(s);
2717 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2718 sp = s + strlen(s);
2719 inet_ntoa_r(inc->inc_laddr, sp);
2720 sp = s + strlen(s);
2721 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2722 #ifdef INET6
2723 } else if (inc) {
2724 ip6_sprintf(sp, &inc->inc6_faddr);
2725 sp = s + strlen(s);
2726 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2727 sp = s + strlen(s);
2728 ip6_sprintf(sp, &inc->inc6_laddr);
2729 sp = s + strlen(s);
2730 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2731 } else if (ip6 && th) {
2732 ip6_sprintf(sp, &ip6->ip6_src);
2733 sp = s + strlen(s);
2734 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2735 sp = s + strlen(s);
2736 ip6_sprintf(sp, &ip6->ip6_dst);
2737 sp = s + strlen(s);
2738 sprintf(sp, "]:%i", ntohs(th->th_dport));
2739 #endif /* INET6 */
2740 #ifdef INET
2741 } else if (ip && th) {
2742 inet_ntoa_r(ip->ip_src, sp);
2743 sp = s + strlen(s);
2744 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2745 sp = s + strlen(s);
2746 inet_ntoa_r(ip->ip_dst, sp);
2747 sp = s + strlen(s);
2748 sprintf(sp, "]:%i", ntohs(th->th_dport));
2749 #endif /* INET */
2750 } else {
2751 free(s, M_TCPLOG);
2752 return (NULL);
2753 }
2754 sp = s + strlen(s);
2755 if (th)
2756 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2757 if (*(s + size - 1) != '\0')
2758 panic("%s: string too long", __func__);
2759 return (s);
2760 }
2761
2762 /*
2763 * A subroutine which makes it easy to track TCP state changes with DTrace.
2764 * This function shouldn't be called for t_state initializations that don't
2765 * correspond to actual TCP state transitions.
2766 */
2767 void
tcp_state_change(struct tcpcb * tp,int newstate)2768 tcp_state_change(struct tcpcb *tp, int newstate)
2769 {
2770 #if defined(KDTRACE_HOOKS)
2771 int pstate = tp->t_state;
2772 #endif
2773
2774 TCPSTATES_DEC(tp->t_state);
2775 TCPSTATES_INC(newstate);
2776 tp->t_state = newstate;
2777 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
2778 }
2779