96#ifdef INET6
97#include <netinet6/tcp6_var.h>
98#endif
99#include <netinet/tcpip.h>
100#ifdef TCPPCAP
101#include <netinet/tcp_pcap.h>
102#endif
103#ifdef TCPDEBUG
104#include <netinet/tcp_debug.h>
105#endif
106#ifdef INET6
107#include <netinet6/ip6protosw.h>
108#endif
109#ifdef TCP_OFFLOAD
110#include <netinet/tcp_offload.h>
111#endif
112
113#ifdef IPSEC
114#include <netipsec/ipsec.h>
115#include <netipsec/xform.h>
116#ifdef INET6
117#include <netipsec/ipsec6.h>
118#endif
119#include <netipsec/key.h>
120#include <sys/syslog.h>
121#endif /*IPSEC*/
122
123#include <machine/in_cksum.h>
124#include <sys/md5.h>
125
126#include <security/mac/mac_framework.h>
127
128VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
129#ifdef INET6
130VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
131#endif
132
133struct rwlock tcp_function_lock;
134
135static int
136sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
137{
138 int error, new;
139
140 new = V_tcp_mssdflt;
141 error = sysctl_handle_int(oidp, &new, 0, req);
142 if (error == 0 && req->newptr) {
143 if (new < TCP_MINMSS)
144 error = EINVAL;
145 else
146 V_tcp_mssdflt = new;
147 }
148 return (error);
149}
150
151SYSCTL_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");
155
156#ifdef INET6
157static int
158sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
159{
160 int error, new;
161
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)
166 error = EINVAL;
167 else
168 V_tcp_v6mssdflt = new;
169 }
170 return (error);
171}
172
173SYSCTL_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");
177#endif /* INET6 */
178
179/*
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.
186 */
187VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
188SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
189 &VNET_NAME(tcp_minmss), 0,
190 "Minimum TCP Maximum Segment Size");
191
192VNET_DEFINE(int, tcp_do_rfc1323) = 1;
193SYSCTL_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");
196
197static int tcp_log_debug = 0;
198SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
199 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
200
201static int tcp_tcbhashsize;
202SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
203 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
204
205static int do_tcpdrain = 1;
206SYSCTL_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");
208
209SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
210 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
211
212static VNET_DEFINE(int, icmp_may_rst) = 1;
213#define V_icmp_may_rst VNET(icmp_may_rst)
214SYSCTL_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");
217
218static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
219#define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
220SYSCTL_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");
223
224static int tcp_soreceive_stream;
225SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
226 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
227
228#ifdef TCP_SIGNATURE
229static int tcp_sig_checksigs = 1;
230SYSCTL_INT(_net_inet_tcp, OID_AUTO, signature_verify_input, CTLFLAG_RW,
231 &tcp_sig_checksigs, 0, "Verify RFC2385 digests on inbound traffic");
232#endif
233
234VNET_DEFINE(uma_zone_t, sack_hole_zone);
235#define V_sack_hole_zone VNET(sack_hole_zone)
236
237VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
238
239static struct inpcb *tcp_notify(struct inpcb *, int);
240static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
241static void tcp_mtudisc(struct inpcb *, int);
242static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
243 void *ip4hdr, const void *ip6hdr);
244static void tcp_timer_discard(struct tcpcb *, uint32_t);
245
246
247static struct tcp_function_block tcp_def_funcblk = {
248 "default",
249 tcp_output,
250 tcp_do_segment,
251 tcp_default_ctloutput,
252 NULL,
253 NULL,
254 NULL,
255 NULL,
256 NULL,
257 NULL,
258 NULL,
259 0,
260 0
261};
262
263struct tcp_funchead t_functions;
264static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
265
266static struct tcp_function_block *
267find_tcp_functions_locked(struct tcp_function_set *fs)
268{
269 struct tcp_function *f;
270 struct tcp_function_block *blk=NULL;
271
272 TAILQ_FOREACH(f, &t_functions, tf_next) {
273 if (strcmp(f->tf_fb->tfb_tcp_block_name, fs->function_set_name) == 0) {
274 blk = f->tf_fb;
275 break;
276 }
277 }
278 return(blk);
279}
280
281static struct tcp_function_block *
282find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
283{
284 struct tcp_function_block *rblk=NULL;
285 struct tcp_function *f;
286
287 TAILQ_FOREACH(f, &t_functions, tf_next) {
288 if (f->tf_fb == blk) {
289 rblk = blk;
290 if (s) {
291 *s = f;
292 }
293 break;
294 }
295 }
296 return (rblk);
297}
298
299struct tcp_function_block *
300find_and_ref_tcp_functions(struct tcp_function_set *fs)
301{
302 struct tcp_function_block *blk;
303
304 rw_rlock(&tcp_function_lock);
305 blk = find_tcp_functions_locked(fs);
306 if (blk)
307 refcount_acquire(&blk->tfb_refcnt);
308 rw_runlock(&tcp_function_lock);
309 return(blk);
310}
311
312struct tcp_function_block *
313find_and_ref_tcp_fb(struct tcp_function_block *blk)
314{
315 struct tcp_function_block *rblk;
316
317 rw_rlock(&tcp_function_lock);
318 rblk = find_tcp_fb_locked(blk, NULL);
319 if (rblk)
320 refcount_acquire(&rblk->tfb_refcnt);
321 rw_runlock(&tcp_function_lock);
322 return(rblk);
323}
324
325
326static int
327sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
328{
329 int error=ENOENT;
330 struct tcp_function_set fs;
331 struct tcp_function_block *blk;
332
333 memset(&fs, 0, sizeof(fs));
334 rw_rlock(&tcp_function_lock);
335 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
336 if (blk) {
337 /* Found him */
338 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
339 fs.pcbcnt = blk->tfb_refcnt;
340 }
341 rw_runlock(&tcp_function_lock);
342 error = sysctl_handle_string(oidp, fs.function_set_name,
343 sizeof(fs.function_set_name), req);
344
345 /* Check for error or no change */
346 if (error != 0 || req->newptr == NULL)
347 return(error);
348
349 rw_wlock(&tcp_function_lock);
350 blk = find_tcp_functions_locked(&fs);
351 if ((blk == NULL) ||
352 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
353 error = ENOENT;
354 goto done;
355 }
356 tcp_func_set_ptr = blk;
357done:
358 rw_wunlock(&tcp_function_lock);
359 return (error);
360}
361
362SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
363 CTLTYPE_STRING | CTLFLAG_RW,
364 NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
365 "Set/get the default TCP functions");
366
367static int
368sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
369{
370 int error, cnt, linesz;
371 struct tcp_function *f;
372 char *buffer, *cp;
373 size_t bufsz, outsz;
374
375 cnt = 0;
376 rw_rlock(&tcp_function_lock);
377 TAILQ_FOREACH(f, &t_functions, tf_next) {
378 cnt++;
379 }
380 rw_runlock(&tcp_function_lock);
381
382 bufsz = (cnt+2) * (TCP_FUNCTION_NAME_LEN_MAX + 12) + 1;
383 buffer = malloc(bufsz, M_TEMP, M_WAITOK);
384
385 error = 0;
386 cp = buffer;
387
388 linesz = snprintf(cp, bufsz, "\n%-32s%c %s\n", "Stack", 'D', "PCB count");
389 cp += linesz;
390 bufsz -= linesz;
391 outsz = linesz;
392
393 rw_rlock(&tcp_function_lock);
394 TAILQ_FOREACH(f, &t_functions, tf_next) {
395 linesz = snprintf(cp, bufsz, "%-32s%c %u\n",
396 f->tf_fb->tfb_tcp_block_name,
397 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
398 f->tf_fb->tfb_refcnt);
399 if (linesz >= bufsz) {
400 error = EOVERFLOW;
401 break;
402 }
403 cp += linesz;
404 bufsz -= linesz;
405 outsz += linesz;
406 }
407 rw_runlock(&tcp_function_lock);
408 if (error == 0)
409 error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
410 free(buffer, M_TEMP);
411 return (error);
412}
413
414SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
415 CTLTYPE_STRING|CTLFLAG_RD,
416 NULL, 0, sysctl_net_inet_list_available, "A",
417 "list available TCP Function sets");
418
419/*
420 * Target size of TCP PCB hash tables. Must be a power of two.
421 *
422 * Note that this can be overridden by the kernel environment
423 * variable net.inet.tcp.tcbhashsize
424 */
425#ifndef TCBHASHSIZE
426#define TCBHASHSIZE 0
427#endif
428
429/*
430 * XXX
431 * Callouts should be moved into struct tcp directly. They are currently
432 * separate because the tcpcb structure is exported to userland for sysctl
433 * parsing purposes, which do not know about callouts.
434 */
435struct tcpcb_mem {
436 struct tcpcb tcb;
437 struct tcp_timer tt;
438 struct cc_var ccv;
439 struct osd osd;
440};
441
442static VNET_DEFINE(uma_zone_t, tcpcb_zone);
443#define V_tcpcb_zone VNET(tcpcb_zone)
444
445MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
446MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
447
448static struct mtx isn_mtx;
449
450#define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
451#define ISN_LOCK() mtx_lock(&isn_mtx)
452#define ISN_UNLOCK() mtx_unlock(&isn_mtx)
453
454/*
455 * TCP initialization.
456 */
457static void
458tcp_zone_change(void *tag)
459{
460
461 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
462 uma_zone_set_max(V_tcpcb_zone, maxsockets);
463 tcp_tw_zone_change();
464}
465
466static int
467tcp_inpcb_init(void *mem, int size, int flags)
468{
469 struct inpcb *inp = mem;
470
471 INP_LOCK_INIT(inp, "inp", "tcpinp");
472 return (0);
473}
474
475/*
476 * Take a value and get the next power of 2 that doesn't overflow.
477 * Used to size the tcp_inpcb hash buckets.
478 */
479static int
480maketcp_hashsize(int size)
481{
482 int hashsize;
483
484 /*
485 * auto tune.
486 * get the next power of 2 higher than maxsockets.
487 */
488 hashsize = 1 << fls(size);
489 /* catch overflow, and just go one power of 2 smaller */
490 if (hashsize < size) {
491 hashsize = 1 << (fls(size) - 1);
492 }
493 return (hashsize);
494}
495
496int
497register_tcp_functions(struct tcp_function_block *blk, int wait)
498{
499 struct tcp_function_block *lblk;
500 struct tcp_function *n;
501 struct tcp_function_set fs;
502
503 if ((blk->tfb_tcp_output == NULL) ||
504 (blk->tfb_tcp_do_segment == NULL) ||
505 (blk->tfb_tcp_ctloutput == NULL) ||
506 (strlen(blk->tfb_tcp_block_name) == 0)) {
507 /*
508 * These functions are required and you
509 * need a name.
510 */
511 return (EINVAL);
512 }
513 if (blk->tfb_tcp_timer_stop_all ||
514 blk->tfb_tcp_timers_left ||
515 blk->tfb_tcp_timer_activate ||
516 blk->tfb_tcp_timer_active ||
517 blk->tfb_tcp_timer_stop) {
518 /*
519 * If you define one timer function you
520 * must have them all.
521 */
522 if ((blk->tfb_tcp_timer_stop_all == NULL) ||
523 (blk->tfb_tcp_timers_left == NULL) ||
524 (blk->tfb_tcp_timer_activate == NULL) ||
525 (blk->tfb_tcp_timer_active == NULL) ||
526 (blk->tfb_tcp_timer_stop == NULL)) {
527 return (EINVAL);
528 }
529 }
530 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
531 if (n == NULL) {
532 return (ENOMEM);
533 }
534 n->tf_fb = blk;
535 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
536 rw_wlock(&tcp_function_lock);
537 lblk = find_tcp_functions_locked(&fs);
538 if (lblk) {
539 /* Duplicate name space not allowed */
540 rw_wunlock(&tcp_function_lock);
541 free(n, M_TCPFUNCTIONS);
542 return (EALREADY);
543 }
544 refcount_init(&blk->tfb_refcnt, 0);
545 blk->tfb_flags = 0;
546 TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
547 rw_wunlock(&tcp_function_lock);
548 return(0);
549}
550
551int
552deregister_tcp_functions(struct tcp_function_block *blk)
553{
554 struct tcp_function_block *lblk;
555 struct tcp_function *f;
556 int error=ENOENT;
557
558 if (strcmp(blk->tfb_tcp_block_name, "default") == 0) {
559 /* You can't un-register the default */
560 return (EPERM);
561 }
562 rw_wlock(&tcp_function_lock);
563 if (blk == tcp_func_set_ptr) {
564 /* You can't free the current default */
565 rw_wunlock(&tcp_function_lock);
566 return (EBUSY);
567 }
568 if (blk->tfb_refcnt) {
569 /* Still tcb attached, mark it. */
570 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
571 rw_wunlock(&tcp_function_lock);
572 return (EBUSY);
573 }
574 lblk = find_tcp_fb_locked(blk, &f);
575 if (lblk) {
576 /* Found */
577 TAILQ_REMOVE(&t_functions, f, tf_next);
578 f->tf_fb = NULL;
579 free(f, M_TCPFUNCTIONS);
580 error = 0;
581 }
582 rw_wunlock(&tcp_function_lock);
583 return (error);
584}
585
586void
587tcp_init(void)
588{
589 const char *tcbhash_tuneable;
590 int hashsize;
591
592 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
593
594 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
595 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
596 printf("%s: WARNING: unable to register helper hook\n", __func__);
597 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
598 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
599 printf("%s: WARNING: unable to register helper hook\n", __func__);
600 hashsize = TCBHASHSIZE;
601 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
602 if (hashsize == 0) {
603 /*
604 * Auto tune the hash size based on maxsockets.
605 * A perfect hash would have a 1:1 mapping
606 * (hashsize = maxsockets) however it's been
607 * suggested that O(2) average is better.
608 */
609 hashsize = maketcp_hashsize(maxsockets / 4);
610 /*
611 * Our historical default is 512,
612 * do not autotune lower than this.
613 */
614 if (hashsize < 512)
615 hashsize = 512;
616 if (bootverbose && IS_DEFAULT_VNET(curvnet))
617 printf("%s: %s auto tuned to %d\n", __func__,
618 tcbhash_tuneable, hashsize);
619 }
620 /*
621 * We require a hashsize to be a power of two.
622 * Previously if it was not a power of two we would just reset it
623 * back to 512, which could be a nasty surprise if you did not notice
624 * the error message.
625 * Instead what we do is clip it to the closest power of two lower
626 * than the specified hash value.
627 */
628 if (!powerof2(hashsize)) {
629 int oldhashsize = hashsize;
630
631 hashsize = maketcp_hashsize(hashsize);
632 /* prevent absurdly low value */
633 if (hashsize < 16)
634 hashsize = 16;
635 printf("%s: WARNING: TCB hash size not a power of 2, "
636 "clipped from %d to %d.\n", __func__, oldhashsize,
637 hashsize);
638 }
639 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
640 "tcp_inpcb", tcp_inpcb_init, NULL, UMA_ZONE_NOFREE,
641 IPI_HASHFIELDS_4TUPLE);
642
643 /*
644 * These have to be type stable for the benefit of the timers.
645 */
646 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
647 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
648 uma_zone_set_max(V_tcpcb_zone, maxsockets);
649 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
650
651 tcp_tw_init();
652 syncache_init();
653 tcp_hc_init();
654
655 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
656 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
657 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
658
659 /* Skip initialization of globals for non-default instances. */
660 if (!IS_DEFAULT_VNET(curvnet))
661 return;
662
663 tcp_reass_global_init();
664
665 /* XXX virtualize those bellow? */
666 tcp_delacktime = TCPTV_DELACK;
667 tcp_keepinit = TCPTV_KEEP_INIT;
668 tcp_keepidle = TCPTV_KEEP_IDLE;
669 tcp_keepintvl = TCPTV_KEEPINTVL;
670 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
671 tcp_msl = TCPTV_MSL;
672 tcp_rexmit_min = TCPTV_MIN;
673 if (tcp_rexmit_min < 1)
674 tcp_rexmit_min = 1;
675 tcp_rexmit_slop = TCPTV_CPU_VAR;
676 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
677 tcp_tcbhashsize = hashsize;
678 /* Setup the tcp function block list */
679 TAILQ_INIT(&t_functions);
680 rw_init_flags(&tcp_function_lock, "tcp_func_lock" , 0);
681 register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
682
683 if (tcp_soreceive_stream) {
684#ifdef INET
685 tcp_usrreqs.pru_soreceive = soreceive_stream;
686#endif
687#ifdef INET6
688 tcp6_usrreqs.pru_soreceive = soreceive_stream;
689#endif /* INET6 */
690 }
691
692#ifdef INET6
693#define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
694#else /* INET6 */
695#define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
696#endif /* INET6 */
697 if (max_protohdr < TCP_MINPROTOHDR)
698 max_protohdr = TCP_MINPROTOHDR;
699 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
700 panic("tcp_init");
701#undef TCP_MINPROTOHDR
702
703 ISN_LOCK_INIT();
704 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
705 SHUTDOWN_PRI_DEFAULT);
706 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
707 EVENTHANDLER_PRI_ANY);
708#ifdef TCPPCAP
709 tcp_pcap_init();
710#endif
711
712#ifdef TCP_RFC7413
713 tcp_fastopen_init();
714#endif
715}
716
717#ifdef VIMAGE
718void
719tcp_destroy(void)
720{
721 int error;
722
723#ifdef TCP_RFC7413
724 tcp_fastopen_destroy();
725#endif
726 tcp_hc_destroy();
727 syncache_destroy();
728 tcp_tw_destroy();
729 in_pcbinfo_destroy(&V_tcbinfo);
730 uma_zdestroy(V_sack_hole_zone);
731 uma_zdestroy(V_tcpcb_zone);
732
733 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
734 if (error != 0) {
735 printf("%s: WARNING: unable to deregister helper hook "
736 "type=%d, id=%d: error %d returned\n", __func__,
737 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
738 }
739 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
740 if (error != 0) {
741 printf("%s: WARNING: unable to deregister helper hook "
742 "type=%d, id=%d: error %d returned\n", __func__,
743 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
744 }
745}
746#endif
747
748void
749tcp_fini(void *xtp)
750{
751
752}
753
754/*
755 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
756 * tcp_template used to store this data in mbufs, but we now recopy it out
757 * of the tcpcb each time to conserve mbufs.
758 */
759void
760tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
761{
762 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
763
764 INP_WLOCK_ASSERT(inp);
765
766#ifdef INET6
767 if ((inp->inp_vflag & INP_IPV6) != 0) {
768 struct ip6_hdr *ip6;
769
770 ip6 = (struct ip6_hdr *)ip_ptr;
771 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
772 (inp->inp_flow & IPV6_FLOWINFO_MASK);
773 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
774 (IPV6_VERSION & IPV6_VERSION_MASK);
775 ip6->ip6_nxt = IPPROTO_TCP;
776 ip6->ip6_plen = htons(sizeof(struct tcphdr));
777 ip6->ip6_src = inp->in6p_laddr;
778 ip6->ip6_dst = inp->in6p_faddr;
779 }
780#endif /* INET6 */
781#if defined(INET6) && defined(INET)
782 else
783#endif
784#ifdef INET
785 {
786 struct ip *ip;
787
788 ip = (struct ip *)ip_ptr;
789 ip->ip_v = IPVERSION;
790 ip->ip_hl = 5;
791 ip->ip_tos = inp->inp_ip_tos;
792 ip->ip_len = 0;
793 ip->ip_id = 0;
794 ip->ip_off = 0;
795 ip->ip_ttl = inp->inp_ip_ttl;
796 ip->ip_sum = 0;
797 ip->ip_p = IPPROTO_TCP;
798 ip->ip_src = inp->inp_laddr;
799 ip->ip_dst = inp->inp_faddr;
800 }
801#endif /* INET */
802 th->th_sport = inp->inp_lport;
803 th->th_dport = inp->inp_fport;
804 th->th_seq = 0;
805 th->th_ack = 0;
806 th->th_x2 = 0;
807 th->th_off = 5;
808 th->th_flags = 0;
809 th->th_win = 0;
810 th->th_urp = 0;
811 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
812}
813
814/*
815 * Create template to be used to send tcp packets on a connection.
816 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
817 * use for this function is in keepalives, which use tcp_respond.
818 */
819struct tcptemp *
820tcpip_maketemplate(struct inpcb *inp)
821{
822 struct tcptemp *t;
823
824 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
825 if (t == NULL)
826 return (NULL);
827 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
828 return (t);
829}
830
831/*
832 * Send a single message to the TCP at address specified by
833 * the given TCP/IP header. If m == NULL, then we make a copy
834 * of the tcpiphdr at th and send directly to the addressed host.
835 * This is used to force keep alive messages out using the TCP
836 * template for a connection. If flags are given then we send
837 * a message back to the TCP which originated the segment th,
838 * and discard the mbuf containing it and any other attached mbufs.
839 *
840 * In any case the ack and sequence number of the transmitted
841 * segment are as specified by the parameters.
842 *
843 * NOTE: If m != NULL, then th must point to *inside* the mbuf.
844 */
845void
846tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
847 tcp_seq ack, tcp_seq seq, int flags)
848{
849 int tlen;
850 int win = 0;
851 struct ip *ip;
852 struct tcphdr *nth;
853#ifdef INET6
854 struct ip6_hdr *ip6;
855 int isipv6;
856#endif /* INET6 */
857 int ipflags = 0;
858 struct inpcb *inp;
859
860 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
861
862#ifdef INET6
863 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
864 ip6 = ipgen;
865#endif /* INET6 */
866 ip = ipgen;
867
868 if (tp != NULL) {
869 inp = tp->t_inpcb;
870 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
871 INP_WLOCK_ASSERT(inp);
872 } else
873 inp = NULL;
874
875 if (tp != NULL) {
876 if (!(flags & TH_RST)) {
877 win = sbspace(&inp->inp_socket->so_rcv);
878 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
879 win = (long)TCP_MAXWIN << tp->rcv_scale;
880 }
881 }
882 if (m == NULL) {
883 m = m_gethdr(M_NOWAIT, MT_DATA);
884 if (m == NULL)
885 return;
886 tlen = 0;
887 m->m_data += max_linkhdr;
888#ifdef INET6
889 if (isipv6) {
890 bcopy((caddr_t)ip6, mtod(m, caddr_t),
891 sizeof(struct ip6_hdr));
892 ip6 = mtod(m, struct ip6_hdr *);
893 nth = (struct tcphdr *)(ip6 + 1);
894 } else
895#endif /* INET6 */
896 {
897 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
898 ip = mtod(m, struct ip *);
899 nth = (struct tcphdr *)(ip + 1);
900 }
901 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
902 flags = TH_ACK;
903 } else {
904 /*
905 * reuse the mbuf.
906 * XXX MRT We inherrit the FIB, which is lucky.
907 */
908 m_freem(m->m_next);
909 m->m_next = NULL;
910 m->m_data = (caddr_t)ipgen;
911 /* m_len is set later */
912 tlen = 0;
913#define xchg(a,b,type) { type t; t=a; a=b; b=t; }
914#ifdef INET6
915 if (isipv6) {
916 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
917 nth = (struct tcphdr *)(ip6 + 1);
918 } else
919#endif /* INET6 */
920 {
921 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
922 nth = (struct tcphdr *)(ip + 1);
923 }
924 if (th != nth) {
925 /*
926 * this is usually a case when an extension header
927 * exists between the IPv6 header and the
928 * TCP header.
929 */
930 nth->th_sport = th->th_sport;
931 nth->th_dport = th->th_dport;
932 }
933 xchg(nth->th_dport, nth->th_sport, uint16_t);
934#undef xchg
935 }
936#ifdef INET6
937 if (isipv6) {
938 ip6->ip6_flow = 0;
939 ip6->ip6_vfc = IPV6_VERSION;
940 ip6->ip6_nxt = IPPROTO_TCP;
941 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
942 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
943 }
944#endif
945#if defined(INET) && defined(INET6)
946 else
947#endif
948#ifdef INET
949 {
950 tlen += sizeof (struct tcpiphdr);
951 ip->ip_len = htons(tlen);
952 ip->ip_ttl = V_ip_defttl;
953 if (V_path_mtu_discovery)
954 ip->ip_off |= htons(IP_DF);
955 }
956#endif
957 m->m_len = tlen;
958 m->m_pkthdr.len = tlen;
959 m->m_pkthdr.rcvif = NULL;
960#ifdef MAC
961 if (inp != NULL) {
962 /*
963 * Packet is associated with a socket, so allow the
964 * label of the response to reflect the socket label.
965 */
966 INP_WLOCK_ASSERT(inp);
967 mac_inpcb_create_mbuf(inp, m);
968 } else {
969 /*
970 * Packet is not associated with a socket, so possibly
971 * update the label in place.
972 */
973 mac_netinet_tcp_reply(m);
974 }
975#endif
976 nth->th_seq = htonl(seq);
977 nth->th_ack = htonl(ack);
978 nth->th_x2 = 0;
979 nth->th_off = sizeof (struct tcphdr) >> 2;
980 nth->th_flags = flags;
981 if (tp != NULL)
982 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
983 else
984 nth->th_win = htons((u_short)win);
985 nth->th_urp = 0;
986
987 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
988#ifdef INET6
989 if (isipv6) {
990 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
991 nth->th_sum = in6_cksum_pseudo(ip6,
992 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
993 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
994 NULL, NULL);
995 }
996#endif /* INET6 */
997#if defined(INET6) && defined(INET)
998 else
999#endif
1000#ifdef INET
1001 {
1002 m->m_pkthdr.csum_flags = CSUM_TCP;
1003 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1004 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1005 }
1006#endif /* INET */
1007#ifdef TCPDEBUG
1008 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1009 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1010#endif
1011 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *));
1012 if (flags & TH_RST)
1013 TCP_PROBE5(accept__refused, NULL, NULL, mtod(m, const char *),
1014 tp, nth);
1015
1016 TCP_PROBE5(send, NULL, tp, mtod(m, const char *), tp, nth);
1017#ifdef INET6
1018 if (isipv6)
1019 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
1020#endif /* INET6 */
1021#if defined(INET) && defined(INET6)
1022 else
1023#endif
1024#ifdef INET
1025 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
1026#endif
1027}
1028
1029/*
1030 * Create a new TCP control block, making an
1031 * empty reassembly queue and hooking it to the argument
1032 * protocol control block. The `inp' parameter must have
1033 * come from the zone allocator set up in tcp_init().
1034 */
1035struct tcpcb *
1036tcp_newtcpcb(struct inpcb *inp)
1037{
1038 struct tcpcb_mem *tm;
1039 struct tcpcb *tp;
1040#ifdef INET6
1041 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1042#endif /* INET6 */
1043
1044 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1045 if (tm == NULL)
1046 return (NULL);
1047 tp = &tm->tcb;
1048
1049 /* Initialise cc_var struct for this tcpcb. */
1050 tp->ccv = &tm->ccv;
1051 tp->ccv->type = IPPROTO_TCP;
1052 tp->ccv->ccvc.tcp = tp;
1053 rw_rlock(&tcp_function_lock);
1054 tp->t_fb = tcp_func_set_ptr;
1055 refcount_acquire(&tp->t_fb->tfb_refcnt);
1056 rw_runlock(&tcp_function_lock);
1057 if (tp->t_fb->tfb_tcp_fb_init) {
1058 (*tp->t_fb->tfb_tcp_fb_init)(tp);
1059 }
1060 /*
1061 * Use the current system default CC algorithm.
1062 */
1063 CC_LIST_RLOCK();
1064 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1065 CC_ALGO(tp) = CC_DEFAULT();
1066 CC_LIST_RUNLOCK();
1067
1068 if (CC_ALGO(tp)->cb_init != NULL)
1069 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1070 if (tp->t_fb->tfb_tcp_fb_fini)
1071 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1072 refcount_release(&tp->t_fb->tfb_refcnt);
1073 uma_zfree(V_tcpcb_zone, tm);
1074 return (NULL);
1075 }
1076
1077 tp->osd = &tm->osd;
1078 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1079 if (tp->t_fb->tfb_tcp_fb_fini)
1080 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1081 refcount_release(&tp->t_fb->tfb_refcnt);
1082 uma_zfree(V_tcpcb_zone, tm);
1083 return (NULL);
1084 }
1085
1086#ifdef VIMAGE
1087 tp->t_vnet = inp->inp_vnet;
1088#endif
1089 tp->t_timers = &tm->tt;
1090 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
1091 tp->t_maxseg =
1092#ifdef INET6
1093 isipv6 ? V_tcp_v6mssdflt :
1094#endif /* INET6 */
1095 V_tcp_mssdflt;
1096
1097 /* Set up our timeouts. */
1098 callout_init(&tp->t_timers->tt_rexmt, 1);
1099 callout_init(&tp->t_timers->tt_persist, 1);
1100 callout_init(&tp->t_timers->tt_keep, 1);
1101 callout_init(&tp->t_timers->tt_2msl, 1);
1102 callout_init(&tp->t_timers->tt_delack, 1);
1103
1104 if (V_tcp_do_rfc1323)
1105 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1106 if (V_tcp_do_sack)
1107 tp->t_flags |= TF_SACK_PERMIT;
1108 TAILQ_INIT(&tp->snd_holes);
1109 /*
1110 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1111 * is called.
1112 */
1113 in_pcbref(inp); /* Reference for tcpcb */
1114 tp->t_inpcb = inp;
1115
1116 /*
1117 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1118 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
1119 * reasonable initial retransmit time.
1120 */
1121 tp->t_srtt = TCPTV_SRTTBASE;
1122 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1123 tp->t_rttmin = tcp_rexmit_min;
1124 tp->t_rxtcur = TCPTV_RTOBASE;
1125 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1126 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1127 tp->t_rcvtime = ticks;
1128 /*
1129 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1130 * because the socket may be bound to an IPv6 wildcard address,
1131 * which may match an IPv4-mapped IPv6 address.
1132 */
1133 inp->inp_ip_ttl = V_ip_defttl;
1134 inp->inp_ppcb = tp;
1135#ifdef TCPPCAP
1136 /*
1137 * Init the TCP PCAP queues.
1138 */
1139 tcp_pcap_tcpcb_init(tp);
1140#endif
1141 return (tp); /* XXX */
1142}
1143
1144/*
1145 * Switch the congestion control algorithm back to NewReno for any active
1146 * control blocks using an algorithm which is about to go away.
1147 * This ensures the CC framework can allow the unload to proceed without leaving
1148 * any dangling pointers which would trigger a panic.
1149 * Returning non-zero would inform the CC framework that something went wrong
1150 * and it would be unsafe to allow the unload to proceed. However, there is no
1151 * way for this to occur with this implementation so we always return zero.
1152 */
1153int
1154tcp_ccalgounload(struct cc_algo *unload_algo)
1155{
1156 struct cc_algo *tmpalgo;
1157 struct inpcb *inp;
1158 struct tcpcb *tp;
1159 VNET_ITERATOR_DECL(vnet_iter);
1160
1161 /*
1162 * Check all active control blocks across all network stacks and change
1163 * any that are using "unload_algo" back to NewReno. If "unload_algo"
1164 * requires cleanup code to be run, call it.
1165 */
1166 VNET_LIST_RLOCK();
1167 VNET_FOREACH(vnet_iter) {
1168 CURVNET_SET(vnet_iter);
1169 INP_INFO_WLOCK(&V_tcbinfo);
1170 /*
1171 * New connections already part way through being initialised
1172 * with the CC algo we're removing will not race with this code
1173 * because the INP_INFO_WLOCK is held during initialisation. We
1174 * therefore don't enter the loop below until the connection
1175 * list has stabilised.
1176 */
1177 LIST_FOREACH(inp, &V_tcb, inp_list) {
1178 INP_WLOCK(inp);
1179 /* Important to skip tcptw structs. */
1180 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1181 (tp = intotcpcb(inp)) != NULL) {
1182 /*
1183 * By holding INP_WLOCK here, we are assured
1184 * that the connection is not currently
1185 * executing inside the CC module's functions
1186 * i.e. it is safe to make the switch back to
1187 * NewReno.
1188 */
1189 if (CC_ALGO(tp) == unload_algo) {
1190 tmpalgo = CC_ALGO(tp);
1191 /* NewReno does not require any init. */
1192 CC_ALGO(tp) = &newreno_cc_algo;
1193 if (tmpalgo->cb_destroy != NULL)
1194 tmpalgo->cb_destroy(tp->ccv);
1195 }
1196 }
1197 INP_WUNLOCK(inp);
1198 }
1199 INP_INFO_WUNLOCK(&V_tcbinfo);
1200 CURVNET_RESTORE();
1201 }
1202 VNET_LIST_RUNLOCK();
1203
1204 return (0);
1205}
1206
1207/*
1208 * Drop a TCP connection, reporting
1209 * the specified error. If connection is synchronized,
1210 * then send a RST to peer.
1211 */
1212struct tcpcb *
1213tcp_drop(struct tcpcb *tp, int errno)
1214{
1215 struct socket *so = tp->t_inpcb->inp_socket;
1216
1217 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1218 INP_WLOCK_ASSERT(tp->t_inpcb);
1219
1220 if (TCPS_HAVERCVDSYN(tp->t_state)) {
1221 tcp_state_change(tp, TCPS_CLOSED);
1222 (void) tp->t_fb->tfb_tcp_output(tp);
1223 TCPSTAT_INC(tcps_drops);
1224 } else
1225 TCPSTAT_INC(tcps_conndrops);
1226 if (errno == ETIMEDOUT && tp->t_softerror)
1227 errno = tp->t_softerror;
1228 so->so_error = errno;
1229 return (tcp_close(tp));
1230}
1231
1232void
1233tcp_discardcb(struct tcpcb *tp)
1234{
1235 struct inpcb *inp = tp->t_inpcb;
1236 struct socket *so = inp->inp_socket;
1237#ifdef INET6
1238 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1239#endif /* INET6 */
1240 int released;
1241
1242 INP_WLOCK_ASSERT(inp);
1243
1244 /*
1245 * Make sure that all of our timers are stopped before we delete the
1246 * PCB.
1247 *
1248 * If stopping a timer fails, we schedule a discard function in same
1249 * callout, and the last discard function called will take care of
1250 * deleting the tcpcb.
1251 */
1252 tcp_timer_stop(tp, TT_REXMT);
1253 tcp_timer_stop(tp, TT_PERSIST);
1254 tcp_timer_stop(tp, TT_KEEP);
1255 tcp_timer_stop(tp, TT_2MSL);
1256 tcp_timer_stop(tp, TT_DELACK);
1257 if (tp->t_fb->tfb_tcp_timer_stop_all) {
1258 /* Call the stop-all function of the methods */
1259 tp->t_fb->tfb_tcp_timer_stop_all(tp);
1260 }
1261
1262 /*
1263 * If we got enough samples through the srtt filter,
1264 * save the rtt and rttvar in the routing entry.
1265 * 'Enough' is arbitrarily defined as 4 rtt samples.
1266 * 4 samples is enough for the srtt filter to converge
1267 * to within enough % of the correct value; fewer samples
1268 * and we could save a bogus rtt. The danger is not high
1269 * as tcp quickly recovers from everything.
1270 * XXX: Works very well but needs some more statistics!
1271 */
1272 if (tp->t_rttupdated >= 4) {
1273 struct hc_metrics_lite metrics;
1274 u_long ssthresh;
1275
1276 bzero(&metrics, sizeof(metrics));
1277 /*
1278 * Update the ssthresh always when the conditions below
1279 * are satisfied. This gives us better new start value
1280 * for the congestion avoidance for new connections.
1281 * ssthresh is only set if packet loss occured on a session.
1282 *
1283 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1284 * being torn down. Ideally this code would not use 'so'.
1285 */
1286 ssthresh = tp->snd_ssthresh;
1287 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1288 /*
1289 * convert the limit from user data bytes to
1290 * packets then to packet data bytes.
1291 */
1292 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1293 if (ssthresh < 2)
1294 ssthresh = 2;
1295 ssthresh *= (u_long)(tp->t_maxseg +
1296#ifdef INET6
1297 (isipv6 ? sizeof (struct ip6_hdr) +
1298 sizeof (struct tcphdr) :
1299#endif
1300 sizeof (struct tcpiphdr)
1301#ifdef INET6
1302 )
1303#endif
1304 );
1305 } else
1306 ssthresh = 0;
1307 metrics.rmx_ssthresh = ssthresh;
1308
1309 metrics.rmx_rtt = tp->t_srtt;
1310 metrics.rmx_rttvar = tp->t_rttvar;
1311 metrics.rmx_cwnd = tp->snd_cwnd;
1312 metrics.rmx_sendpipe = 0;
1313 metrics.rmx_recvpipe = 0;
1314
1315 tcp_hc_update(&inp->inp_inc, &metrics);
1316 }
1317
1318 /* free the reassembly queue, if any */
1319 tcp_reass_flush(tp);
1320
1321#ifdef TCP_OFFLOAD
1322 /* Disconnect offload device, if any. */
1323 if (tp->t_flags & TF_TOE)
1324 tcp_offload_detach(tp);
1325#endif
1326
1327 tcp_free_sackholes(tp);
1328
1329#ifdef TCPPCAP
1330 /* Free the TCP PCAP queues. */
1331 tcp_pcap_drain(&(tp->t_inpkts));
1332 tcp_pcap_drain(&(tp->t_outpkts));
1333#endif
1334
1335 /* Allow the CC algorithm to clean up after itself. */
1336 if (CC_ALGO(tp)->cb_destroy != NULL)
1337 CC_ALGO(tp)->cb_destroy(tp->ccv);
1338
1339 khelp_destroy_osd(tp->osd);
1340
1341 CC_ALGO(tp) = NULL;
1342 inp->inp_ppcb = NULL;
1343 if ((tp->t_timers->tt_flags & TT_MASK) == 0) {
1344 /* We own the last reference on tcpcb, let's free it. */
1345 if ((tp->t_fb->tfb_tcp_timers_left) &&
1346 (tp->t_fb->tfb_tcp_timers_left(tp))) {
1347 /* Some fb timers left running! */
1348 return;
1349 }
1350 if (tp->t_fb->tfb_tcp_fb_fini)
1351 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1352 refcount_release(&tp->t_fb->tfb_refcnt);
1353 tp->t_inpcb = NULL;
1354 uma_zfree(V_tcpcb_zone, tp);
1355 released = in_pcbrele_wlocked(inp);
1356 KASSERT(!released, ("%s: inp %p should not have been released "
1357 "here", __func__, inp));
1358 }
1359}
1360
1361void
1362tcp_timer_2msl_discard(void *xtp)
1363{
1364
1365 tcp_timer_discard((struct tcpcb *)xtp, TT_2MSL);
1366}
1367
1368void
1369tcp_timer_keep_discard(void *xtp)
1370{
1371
1372 tcp_timer_discard((struct tcpcb *)xtp, TT_KEEP);
1373}
1374
1375void
1376tcp_timer_persist_discard(void *xtp)
1377{
1378
1379 tcp_timer_discard((struct tcpcb *)xtp, TT_PERSIST);
1380}
1381
1382void
1383tcp_timer_rexmt_discard(void *xtp)
1384{
1385
1386 tcp_timer_discard((struct tcpcb *)xtp, TT_REXMT);
1387}
1388
1389void
1390tcp_timer_delack_discard(void *xtp)
1391{
1392
1393 tcp_timer_discard((struct tcpcb *)xtp, TT_DELACK);
1394}
1395
1396void
1397tcp_timer_discard(struct tcpcb *tp, uint32_t timer_type)
1398{
1399 struct inpcb *inp;
1400
1401 CURVNET_SET(tp->t_vnet);
1402 INP_INFO_RLOCK(&V_tcbinfo);
1403 inp = tp->t_inpcb;
1404 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1405 __func__, tp));
1406 INP_WLOCK(inp);
1407 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1408 ("%s: tcpcb has to be stopped here", __func__));
1409 KASSERT((tp->t_timers->tt_flags & timer_type) != 0,
1410 ("%s: discard callout should be running", __func__));
1411 tp->t_timers->tt_flags &= ~timer_type;
1412 if ((tp->t_timers->tt_flags & TT_MASK) == 0) {
1413 /* We own the last reference on this tcpcb, let's free it. */
1414 if ((tp->t_fb->tfb_tcp_timers_left) &&
1415 (tp->t_fb->tfb_tcp_timers_left(tp))) {
1416 /* Some fb timers left running! */
1417 goto leave;
1418 }
1419 if (tp->t_fb->tfb_tcp_fb_fini)
1420 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1421 refcount_release(&tp->t_fb->tfb_refcnt);
1422 tp->t_inpcb = NULL;
1423 uma_zfree(V_tcpcb_zone, tp);
1424 if (in_pcbrele_wlocked(inp)) {
1425 INP_INFO_RUNLOCK(&V_tcbinfo);
1426 CURVNET_RESTORE();
1427 return;
1428 }
1429 }
1430leave:
1431 INP_WUNLOCK(inp);
1432 INP_INFO_RUNLOCK(&V_tcbinfo);
1433 CURVNET_RESTORE();
1434}
1435
1436/*
1437 * Attempt to close a TCP control block, marking it as dropped, and freeing
1438 * the socket if we hold the only reference.
1439 */
1440struct tcpcb *
1441tcp_close(struct tcpcb *tp)
1442{
1443 struct inpcb *inp = tp->t_inpcb;
1444 struct socket *so;
1445
1446 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1447 INP_WLOCK_ASSERT(inp);
1448
1449#ifdef TCP_OFFLOAD
1450 if (tp->t_state == TCPS_LISTEN)
1451 tcp_offload_listen_stop(tp);
1452#endif
1453#ifdef TCP_RFC7413
1454 /*
1455 * This releases the TFO pending counter resource for TFO listen
1456 * sockets as well as passively-created TFO sockets that transition
1457 * from SYN_RECEIVED to CLOSED.
1458 */
1459 if (tp->t_tfo_pending) {
1460 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
1461 tp->t_tfo_pending = NULL;
1462 }
1463#endif
1464 in_pcbdrop(inp);
1465 TCPSTAT_INC(tcps_closed);
1466 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1467 so = inp->inp_socket;
1468 soisdisconnected(so);
1469 if (inp->inp_flags & INP_SOCKREF) {
1470 KASSERT(so->so_state & SS_PROTOREF,
1471 ("tcp_close: !SS_PROTOREF"));
1472 inp->inp_flags &= ~INP_SOCKREF;
1473 INP_WUNLOCK(inp);
1474 ACCEPT_LOCK();
1475 SOCK_LOCK(so);
1476 so->so_state &= ~SS_PROTOREF;
1477 sofree(so);
1478 return (NULL);
1479 }
1480 return (tp);
1481}
1482
1483void
1484tcp_drain(void)
1485{
1486 VNET_ITERATOR_DECL(vnet_iter);
1487
1488 if (!do_tcpdrain)
1489 return;
1490
1491 VNET_LIST_RLOCK_NOSLEEP();
1492 VNET_FOREACH(vnet_iter) {
1493 CURVNET_SET(vnet_iter);
1494 struct inpcb *inpb;
1495 struct tcpcb *tcpb;
1496
1497 /*
1498 * Walk the tcpbs, if existing, and flush the reassembly queue,
1499 * if there is one...
1500 * XXX: The "Net/3" implementation doesn't imply that the TCP
1501 * reassembly queue should be flushed, but in a situation
1502 * where we're really low on mbufs, this is potentially
1503 * useful.
1504 */
1505 INP_INFO_WLOCK(&V_tcbinfo);
1506 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1507 if (inpb->inp_flags & INP_TIMEWAIT)
1508 continue;
1509 INP_WLOCK(inpb);
1510 if ((tcpb = intotcpcb(inpb)) != NULL) {
1511 tcp_reass_flush(tcpb);
1512 tcp_clean_sackreport(tcpb);
1513 }
1514 INP_WUNLOCK(inpb);
1515 }
1516 INP_INFO_WUNLOCK(&V_tcbinfo);
1517 CURVNET_RESTORE();
1518 }
1519 VNET_LIST_RUNLOCK_NOSLEEP();
1520}
1521
1522/*
1523 * Notify a tcp user of an asynchronous error;
1524 * store error as soft error, but wake up user
1525 * (for now, won't do anything until can select for soft error).
1526 *
1527 * Do not wake up user since there currently is no mechanism for
1528 * reporting soft errors (yet - a kqueue filter may be added).
1529 */
1530static struct inpcb *
1531tcp_notify(struct inpcb *inp, int error)
1532{
1533 struct tcpcb *tp;
1534
1535 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1536 INP_WLOCK_ASSERT(inp);
1537
1538 if ((inp->inp_flags & INP_TIMEWAIT) ||
1539 (inp->inp_flags & INP_DROPPED))
1540 return (inp);
1541
1542 tp = intotcpcb(inp);
1543 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1544
1545 /*
1546 * Ignore some errors if we are hooked up.
1547 * If connection hasn't completed, has retransmitted several times,
1548 * and receives a second error, give up now. This is better
1549 * than waiting a long time to establish a connection that
1550 * can never complete.
1551 */
1552 if (tp->t_state == TCPS_ESTABLISHED &&
1553 (error == EHOSTUNREACH || error == ENETUNREACH ||
1554 error == EHOSTDOWN)) {
1555 return (inp);
1556 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1557 tp->t_softerror) {
1558 tp = tcp_drop(tp, error);
1559 if (tp != NULL)
1560 return (inp);
1561 else
1562 return (NULL);
1563 } else {
1564 tp->t_softerror = error;
1565 return (inp);
1566 }
1567#if 0
1568 wakeup( &so->so_timeo);
1569 sorwakeup(so);
1570 sowwakeup(so);
1571#endif
1572}
1573
1574static int
1575tcp_pcblist(SYSCTL_HANDLER_ARGS)
1576{
1577 int error, i, m, n, pcb_count;
1578 struct inpcb *inp, **inp_list;
1579 inp_gen_t gencnt;
1580 struct xinpgen xig;
1581
1582 /*
1583 * The process of preparing the TCB list is too time-consuming and
1584 * resource-intensive to repeat twice on every request.
1585 */
1586 if (req->oldptr == NULL) {
1587 n = V_tcbinfo.ipi_count + syncache_pcbcount();
1588 n += imax(n / 8, 10);
1589 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1590 return (0);
1591 }
1592
1593 if (req->newptr != NULL)
1594 return (EPERM);
1595
1596 /*
1597 * OK, now we're committed to doing something.
1598 */
1599 INP_LIST_RLOCK(&V_tcbinfo);
1600 gencnt = V_tcbinfo.ipi_gencnt;
1601 n = V_tcbinfo.ipi_count;
1602 INP_LIST_RUNLOCK(&V_tcbinfo);
1603
1604 m = syncache_pcbcount();
1605
1606 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1607 + (n + m) * sizeof(struct xtcpcb));
1608 if (error != 0)
1609 return (error);
1610
1611 xig.xig_len = sizeof xig;
1612 xig.xig_count = n + m;
1613 xig.xig_gen = gencnt;
1614 xig.xig_sogen = so_gencnt;
1615 error = SYSCTL_OUT(req, &xig, sizeof xig);
1616 if (error)
1617 return (error);
1618
1619 error = syncache_pcblist(req, m, &pcb_count);
1620 if (error)
1621 return (error);
1622
1623 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1624 if (inp_list == NULL)
1625 return (ENOMEM);
1626
1627 INP_INFO_WLOCK(&V_tcbinfo);
1628 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1629 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1630 INP_WLOCK(inp);
1631 if (inp->inp_gencnt <= gencnt) {
1632 /*
1633 * XXX: This use of cr_cansee(), introduced with
1634 * TCP state changes, is not quite right, but for
1635 * now, better than nothing.
1636 */
1637 if (inp->inp_flags & INP_TIMEWAIT) {
1638 if (intotw(inp) != NULL)
1639 error = cr_cansee(req->td->td_ucred,
1640 intotw(inp)->tw_cred);
1641 else
1642 error = EINVAL; /* Skip this inp. */
1643 } else
1644 error = cr_canseeinpcb(req->td->td_ucred, inp);
1645 if (error == 0) {
1646 in_pcbref(inp);
1647 inp_list[i++] = inp;
1648 }
1649 }
1650 INP_WUNLOCK(inp);
1651 }
1652 INP_INFO_WUNLOCK(&V_tcbinfo);
1653 n = i;
1654
1655 error = 0;
1656 for (i = 0; i < n; i++) {
1657 inp = inp_list[i];
1658 INP_RLOCK(inp);
1659 if (inp->inp_gencnt <= gencnt) {
1660 struct xtcpcb xt;
1661 void *inp_ppcb;
1662
1663 bzero(&xt, sizeof(xt));
1664 xt.xt_len = sizeof xt;
1665 /* XXX should avoid extra copy */
1666 bcopy(inp, &xt.xt_inp, sizeof *inp);
1667 inp_ppcb = inp->inp_ppcb;
1668 if (inp_ppcb == NULL)
1669 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1670 else if (inp->inp_flags & INP_TIMEWAIT) {
1671 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1672 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1673 } else {
1674 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1675 if (xt.xt_tp.t_timers)
1676 tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1677 }
1678 if (inp->inp_socket != NULL)
1679 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1680 else {
1681 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1682 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1683 }
1684 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1685 INP_RUNLOCK(inp);
1686 error = SYSCTL_OUT(req, &xt, sizeof xt);
1687 } else
1688 INP_RUNLOCK(inp);
1689 }
1690 INP_INFO_RLOCK(&V_tcbinfo);
1691 for (i = 0; i < n; i++) {
1692 inp = inp_list[i];
1693 INP_RLOCK(inp);
1694 if (!in_pcbrele_rlocked(inp))
1695 INP_RUNLOCK(inp);
1696 }
1697 INP_INFO_RUNLOCK(&V_tcbinfo);
1698
1699 if (!error) {
1700 /*
1701 * Give the user an updated idea of our state.
1702 * If the generation differs from what we told
1703 * her before, she knows that something happened
1704 * while we were processing this request, and it
1705 * might be necessary to retry.
1706 */
1707 INP_LIST_RLOCK(&V_tcbinfo);
1708 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1709 xig.xig_sogen = so_gencnt;
1710 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1711 INP_LIST_RUNLOCK(&V_tcbinfo);
1712 error = SYSCTL_OUT(req, &xig, sizeof xig);
1713 }
1714 free(inp_list, M_TEMP);
1715 return (error);
1716}
1717
1718SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1719 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1720 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1721
1722#ifdef INET
1723static int
1724tcp_getcred(SYSCTL_HANDLER_ARGS)
1725{
1726 struct xucred xuc;
1727 struct sockaddr_in addrs[2];
1728 struct inpcb *inp;
1729 int error;
1730
1731 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1732 if (error)
1733 return (error);
1734 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1735 if (error)
1736 return (error);
1737 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1738 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1739 if (inp != NULL) {
1740 if (inp->inp_socket == NULL)
1741 error = ENOENT;
1742 if (error == 0)
1743 error = cr_canseeinpcb(req->td->td_ucred, inp);
1744 if (error == 0)
1745 cru2x(inp->inp_cred, &xuc);
1746 INP_RUNLOCK(inp);
1747 } else
1748 error = ENOENT;
1749 if (error == 0)
1750 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1751 return (error);
1752}
1753
1754SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1755 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1756 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1757#endif /* INET */
1758
1759#ifdef INET6
1760static int
1761tcp6_getcred(SYSCTL_HANDLER_ARGS)
1762{
1763 struct xucred xuc;
1764 struct sockaddr_in6 addrs[2];
1765 struct inpcb *inp;
1766 int error;
1767#ifdef INET
1768 int mapped = 0;
1769#endif
1770
1771 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1772 if (error)
1773 return (error);
1774 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1775 if (error)
1776 return (error);
1777 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1778 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1779 return (error);
1780 }
1781 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1782#ifdef INET
1783 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1784 mapped = 1;
1785 else
1786#endif
1787 return (EINVAL);
1788 }
1789
1790#ifdef INET
1791 if (mapped == 1)
1792 inp = in_pcblookup(&V_tcbinfo,
1793 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1794 addrs[1].sin6_port,
1795 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1796 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1797 else
1798#endif
1799 inp = in6_pcblookup(&V_tcbinfo,
1800 &addrs[1].sin6_addr, addrs[1].sin6_port,
1801 &addrs[0].sin6_addr, addrs[0].sin6_port,
1802 INPLOOKUP_RLOCKPCB, NULL);
1803 if (inp != NULL) {
1804 if (inp->inp_socket == NULL)
1805 error = ENOENT;
1806 if (error == 0)
1807 error = cr_canseeinpcb(req->td->td_ucred, inp);
1808 if (error == 0)
1809 cru2x(inp->inp_cred, &xuc);
1810 INP_RUNLOCK(inp);
1811 } else
1812 error = ENOENT;
1813 if (error == 0)
1814 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1815 return (error);
1816}
1817
1818SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1819 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1820 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1821#endif /* INET6 */
1822
1823
1824#ifdef INET
1825void
1826tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1827{
1828 struct ip *ip = vip;
1829 struct tcphdr *th;
1830 struct in_addr faddr;
1831 struct inpcb *inp;
1832 struct tcpcb *tp;
1833 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1834 struct icmp *icp;
1835 struct in_conninfo inc;
1836 tcp_seq icmp_tcp_seq;
1837 int mtu;
1838
1839 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1840 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1841 return;
1842
1843 if (cmd == PRC_MSGSIZE)
1844 notify = tcp_mtudisc_notify;
1845 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1846 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1847 notify = tcp_drop_syn_sent;
1848 /*
1849 * Redirects don't need to be handled up here.
1850 */
1851 else if (PRC_IS_REDIRECT(cmd))
1852 return;
1853 /*
1854 * Hostdead is ugly because it goes linearly through all PCBs.
1855 * XXX: We never get this from ICMP, otherwise it makes an
1856 * excellent DoS attack on machines with many connections.
1857 */
1858 else if (cmd == PRC_HOSTDEAD)
1859 ip = NULL;
1860 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1861 return;
1862
1863 if (ip == NULL) {
1864 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1865 return;
1866 }
1867
1868 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
1869 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1870 INP_INFO_RLOCK(&V_tcbinfo);
1871 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
1872 th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1873 if (inp != NULL) {
1874 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1875 !(inp->inp_flags & INP_DROPPED) &&
1876 !(inp->inp_socket == NULL)) {
1877 icmp_tcp_seq = ntohl(th->th_seq);
1878 tp = intotcpcb(inp);
1879 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1880 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1881 if (cmd == PRC_MSGSIZE) {
1882 /*
1883 * MTU discovery:
1884 * If we got a needfrag set the MTU
1885 * in the route to the suggested new
1886 * value (if given) and then notify.
1887 */
1888 mtu = ntohs(icp->icmp_nextmtu);
1889 /*
1890 * If no alternative MTU was
1891 * proposed, try the next smaller
1892 * one.
1893 */
1894 if (!mtu)
1895 mtu = ip_next_mtu(
1896 ntohs(ip->ip_len), 1);
1897 if (mtu < V_tcp_minmss +
1898 sizeof(struct tcpiphdr))
1899 mtu = V_tcp_minmss +
1900 sizeof(struct tcpiphdr);
1901 /*
1902 * Only process the offered MTU if it
1903 * is smaller than the current one.
1904 */
1905 if (mtu < tp->t_maxseg +
1906 sizeof(struct tcpiphdr)) {
1907 bzero(&inc, sizeof(inc));
1908 inc.inc_faddr = faddr;
1909 inc.inc_fibnum =
1910 inp->inp_inc.inc_fibnum;
1911 tcp_hc_updatemtu(&inc, mtu);
1912 tcp_mtudisc(inp, mtu);
1913 }
1914 } else
1915 inp = (*notify)(inp,
1916 inetctlerrmap[cmd]);
1917 }
1918 }
1919 if (inp != NULL)
1920 INP_WUNLOCK(inp);
1921 } else {
1922 bzero(&inc, sizeof(inc));
1923 inc.inc_fport = th->th_dport;
1924 inc.inc_lport = th->th_sport;
1925 inc.inc_faddr = faddr;
1926 inc.inc_laddr = ip->ip_src;
1927 syncache_unreach(&inc, th);
1928 }
1929 INP_INFO_RUNLOCK(&V_tcbinfo);
1930}
1931#endif /* INET */
1932
1933#ifdef INET6
1934void
1935tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1936{
1937 struct tcphdr th;
1938 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1939 struct ip6_hdr *ip6;
1940 struct mbuf *m;
1941 struct ip6ctlparam *ip6cp = NULL;
1942 const struct sockaddr_in6 *sa6_src = NULL;
1943 int off;
1944 struct tcp_portonly {
1945 u_int16_t th_sport;
1946 u_int16_t th_dport;
1947 } *thp;
1948
1949 if (sa->sa_family != AF_INET6 ||
1950 sa->sa_len != sizeof(struct sockaddr_in6))
1951 return;
1952
1953 if (cmd == PRC_MSGSIZE)
1954 notify = tcp_mtudisc_notify;
1955 else if (!PRC_IS_REDIRECT(cmd) &&
1956 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1957 return;
1958
1959 /* if the parameter is from icmp6, decode it. */
1960 if (d != NULL) {
1961 ip6cp = (struct ip6ctlparam *)d;
1962 m = ip6cp->ip6c_m;
1963 ip6 = ip6cp->ip6c_ip6;
1964 off = ip6cp->ip6c_off;
1965 sa6_src = ip6cp->ip6c_src;
1966 } else {
1967 m = NULL;
1968 ip6 = NULL;
1969 off = 0; /* fool gcc */
1970 sa6_src = &sa6_any;
1971 }
1972
1973 if (ip6 != NULL) {
1974 struct in_conninfo inc;
1975 /*
1976 * XXX: We assume that when IPV6 is non NULL,
1977 * M and OFF are valid.
1978 */
1979
1980 /* check if we can safely examine src and dst ports */
1981 if (m->m_pkthdr.len < off + sizeof(*thp))
1982 return;
1983
1984 bzero(&th, sizeof(th));
1985 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1986
1987 in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1988 (struct sockaddr *)ip6cp->ip6c_src,
1989 th.th_sport, cmd, NULL, notify);
1990
1991 bzero(&inc, sizeof(inc));
1992 inc.inc_fport = th.th_dport;
1993 inc.inc_lport = th.th_sport;
1994 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1995 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1996 inc.inc_flags |= INC_ISIPV6;
1997 INP_INFO_RLOCK(&V_tcbinfo);
1998 syncache_unreach(&inc, &th);
1999 INP_INFO_RUNLOCK(&V_tcbinfo);
2000 } else
2001 in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
2002 0, cmd, NULL, notify);
2003}
2004#endif /* INET6 */
2005
2006
2007/*
2008 * Following is where TCP initial sequence number generation occurs.
2009 *
2010 * There are two places where we must use initial sequence numbers:
2011 * 1. In SYN-ACK packets.
2012 * 2. In SYN packets.
2013 *
2014 * All ISNs for SYN-ACK packets are generated by the syncache. See
2015 * tcp_syncache.c for details.
2016 *
2017 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2018 * depends on this property. In addition, these ISNs should be
2019 * unguessable so as to prevent connection hijacking. To satisfy
2020 * the requirements of this situation, the algorithm outlined in
2021 * RFC 1948 is used, with only small modifications.
2022 *
2023 * Implementation details:
2024 *
2025 * Time is based off the system timer, and is corrected so that it
2026 * increases by one megabyte per second. This allows for proper
2027 * recycling on high speed LANs while still leaving over an hour
2028 * before rollover.
2029 *
2030 * As reading the *exact* system time is too expensive to be done
2031 * whenever setting up a TCP connection, we increment the time
2032 * offset in two ways. First, a small random positive increment
2033 * is added to isn_offset for each connection that is set up.
2034 * Second, the function tcp_isn_tick fires once per clock tick
2035 * and increments isn_offset as necessary so that sequence numbers
2036 * are incremented at approximately ISN_BYTES_PER_SECOND. The
2037 * random positive increments serve only to ensure that the same
2038 * exact sequence number is never sent out twice (as could otherwise
2039 * happen when a port is recycled in less than the system tick
2040 * interval.)
2041 *
2042 * net.inet.tcp.isn_reseed_interval controls the number of seconds
2043 * between seeding of isn_secret. This is normally set to zero,
2044 * as reseeding should not be necessary.
2045 *
2046 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2047 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
2048 * general, this means holding an exclusive (write) lock.
2049 */
2050
2051#define ISN_BYTES_PER_SECOND 1048576
2052#define ISN_STATIC_INCREMENT 4096
2053#define ISN_RANDOM_INCREMENT (4096 - 1)
2054
2055static VNET_DEFINE(u_char, isn_secret[32]);
2056static VNET_DEFINE(int, isn_last);
2057static VNET_DEFINE(int, isn_last_reseed);
2058static VNET_DEFINE(u_int32_t, isn_offset);
2059static VNET_DEFINE(u_int32_t, isn_offset_old);
2060
2061#define V_isn_secret VNET(isn_secret)
2062#define V_isn_last VNET(isn_last)
2063#define V_isn_last_reseed VNET(isn_last_reseed)
2064#define V_isn_offset VNET(isn_offset)
2065#define V_isn_offset_old VNET(isn_offset_old)
2066
2067tcp_seq
2068tcp_new_isn(struct tcpcb *tp)
2069{
2070 MD5_CTX isn_ctx;
2071 u_int32_t md5_buffer[4];
2072 tcp_seq new_isn;
2073 u_int32_t projected_offset;
2074
2075 INP_WLOCK_ASSERT(tp->t_inpcb);
2076
2077 ISN_LOCK();
2078 /* Seed if this is the first use, reseed if requested. */
2079 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2080 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2081 < (u_int)ticks))) {
2082 read_random(&V_isn_secret, sizeof(V_isn_secret));
2083 V_isn_last_reseed = ticks;
2084 }
2085
2086 /* Compute the md5 hash and return the ISN. */
2087 MD5Init(&isn_ctx);
2088 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
2089 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
2090#ifdef INET6
2091 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
2092 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
2093 sizeof(struct in6_addr));
2094 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
2095 sizeof(struct in6_addr));
2096 } else
2097#endif
2098 {
2099 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
2100 sizeof(struct in_addr));
2101 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
2102 sizeof(struct in_addr));
2103 }
2104 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
2105 MD5Final((u_char *) &md5_buffer, &isn_ctx);
2106 new_isn = (tcp_seq) md5_buffer[0];
2107 V_isn_offset += ISN_STATIC_INCREMENT +
2108 (arc4random() & ISN_RANDOM_INCREMENT);
2109 if (ticks != V_isn_last) {
2110 projected_offset = V_isn_offset_old +
2111 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2112 if (SEQ_GT(projected_offset, V_isn_offset))
2113 V_isn_offset = projected_offset;
2114 V_isn_offset_old = V_isn_offset;
2115 V_isn_last = ticks;
2116 }
2117 new_isn += V_isn_offset;
2118 ISN_UNLOCK();
2119 return (new_isn);
2120}
2121
2122/*
2123 * When a specific ICMP unreachable message is received and the
2124 * connection state is SYN-SENT, drop the connection. This behavior
2125 * is controlled by the icmp_may_rst sysctl.
2126 */
2127struct inpcb *
2128tcp_drop_syn_sent(struct inpcb *inp, int errno)
2129{
2130 struct tcpcb *tp;
2131
2132 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2133 INP_WLOCK_ASSERT(inp);
2134
2135 if ((inp->inp_flags & INP_TIMEWAIT) ||
2136 (inp->inp_flags & INP_DROPPED))
2137 return (inp);
2138
2139 tp = intotcpcb(inp);
2140 if (tp->t_state != TCPS_SYN_SENT)
2141 return (inp);
2142
2143 tp = tcp_drop(tp, errno);
2144 if (tp != NULL)
2145 return (inp);
2146 else
2147 return (NULL);
2148}
2149
2150/*
2151 * When `need fragmentation' ICMP is received, update our idea of the MSS
2152 * based on the new value. Also nudge TCP to send something, since we
2153 * know the packet we just sent was dropped.
2154 * This duplicates some code in the tcp_mss() function in tcp_input.c.
2155 */
2156static struct inpcb *
2157tcp_mtudisc_notify(struct inpcb *inp, int error)
2158{
2159
2160 tcp_mtudisc(inp, -1);
2161 return (inp);
2162}
2163
2164static void
2165tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2166{
2167 struct tcpcb *tp;
2168 struct socket *so;
2169
2170 INP_WLOCK_ASSERT(inp);
2171 if ((inp->inp_flags & INP_TIMEWAIT) ||
2172 (inp->inp_flags & INP_DROPPED))
2173 return;
2174
2175 tp = intotcpcb(inp);
2176 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2177
2178 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2179
2180 so = inp->inp_socket;
2181 SOCKBUF_LOCK(&so->so_snd);
2182 /* If the mss is larger than the socket buffer, decrease the mss. */
2183 if (so->so_snd.sb_hiwat < tp->t_maxseg)
2184 tp->t_maxseg = so->so_snd.sb_hiwat;
2185 SOCKBUF_UNLOCK(&so->so_snd);
2186
2187 TCPSTAT_INC(tcps_mturesent);
2188 tp->t_rtttime = 0;
2189 tp->snd_nxt = tp->snd_una;
2190 tcp_free_sackholes(tp);
2191 tp->snd_recover = tp->snd_max;
2192 if (tp->t_flags & TF_SACK_PERMIT)
2193 EXIT_FASTRECOVERY(tp->t_flags);
2194 tp->t_fb->tfb_tcp_output(tp);
2195}
2196
2197#ifdef INET
2198/*
2199 * Look-up the routing entry to the peer of this inpcb. If no route
2200 * is found and it cannot be allocated, then return 0. This routine
2201 * is called by TCP routines that access the rmx structure and by
2202 * tcp_mss_update to get the peer/interface MTU.
2203 */
2204u_long
2205tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2206{
2207 struct route sro;
2208 struct sockaddr_in *dst;
2209 struct ifnet *ifp;
2210 u_long maxmtu = 0;
2211
2212 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2213
2214 bzero(&sro, sizeof(sro));
2215 if (inc->inc_faddr.s_addr != INADDR_ANY) {
2216 dst = (struct sockaddr_in *)&sro.ro_dst;
2217 dst->sin_family = AF_INET;
2218 dst->sin_len = sizeof(*dst);
2219 dst->sin_addr = inc->inc_faddr;
2220 in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
2221 }
2222 if (sro.ro_rt != NULL) {
2223 ifp = sro.ro_rt->rt_ifp;
2224 if (sro.ro_rt->rt_mtu == 0)
2225 maxmtu = ifp->if_mtu;
2226 else
2227 maxmtu = min(sro.ro_rt->rt_mtu, ifp->if_mtu);
2228
2229 /* Report additional interface capabilities. */
2230 if (cap != NULL) {
2231 if (ifp->if_capenable & IFCAP_TSO4 &&
2232 ifp->if_hwassist & CSUM_TSO) {
2233 cap->ifcap |= CSUM_TSO;
2234 cap->tsomax = ifp->if_hw_tsomax;
2235 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2236 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2237 }
2238 }
2239 RTFREE(sro.ro_rt);
2240 }
2241 return (maxmtu);
2242}
2243#endif /* INET */
2244
2245#ifdef INET6
2246u_long
2247tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2248{
2249 struct route_in6 sro6;
2250 struct ifnet *ifp;
2251 u_long maxmtu = 0;
2252
2253 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2254
2255 bzero(&sro6, sizeof(sro6));
2256 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2257 sro6.ro_dst.sin6_family = AF_INET6;
2258 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
2259 sro6.ro_dst.sin6_addr = inc->inc6_faddr;
2260 in6_rtalloc_ign(&sro6, 0, inc->inc_fibnum);
2261 }
2262 if (sro6.ro_rt != NULL) {
2263 ifp = sro6.ro_rt->rt_ifp;
2264 if (sro6.ro_rt->rt_mtu == 0)
2265 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
2266 else
2267 maxmtu = min(sro6.ro_rt->rt_mtu,
2268 IN6_LINKMTU(sro6.ro_rt->rt_ifp));
2269
2270 /* Report additional interface capabilities. */
2271 if (cap != NULL) {
2272 if (ifp->if_capenable & IFCAP_TSO6 &&
2273 ifp->if_hwassist & CSUM_TSO) {
2274 cap->ifcap |= CSUM_TSO;
2275 cap->tsomax = ifp->if_hw_tsomax;
2276 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2277 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2278 }
2279 }
2280 RTFREE(sro6.ro_rt);
2281 }
2282
2283 return (maxmtu);
2284}
2285#endif /* INET6 */
2286
2287/*
2288 * Calculate effective SMSS per RFC5681 definition for a given TCP
2289 * connection at its current state, taking into account SACK and etc.
2290 */
2291u_int
2292tcp_maxseg(const struct tcpcb *tp)
2293{
2294 u_int optlen;
2295
2296 if (tp->t_flags & TF_NOOPT)
2297 return (tp->t_maxseg);
2298
2299 /*
2300 * Here we have a simplified code from tcp_addoptions(),
2301 * without a proper loop, and having most of paddings hardcoded.
2302 * We might make mistakes with padding here in some edge cases,
2303 * but this is harmless, since result of tcp_maxseg() is used
2304 * only in cwnd and ssthresh estimations.
2305 */
2306#define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4)
2307 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2308 if (tp->t_flags & TF_RCVD_TSTMP)
2309 optlen = TCPOLEN_TSTAMP_APPA;
2310 else
2311 optlen = 0;
2312#ifdef TCP_SIGNATURE
2313 if (tp->t_flags & TF_SIGNATURE)
2314 optlen += PAD(TCPOLEN_SIGNATURE);
2315#endif
2316 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
2317 optlen += TCPOLEN_SACKHDR;
2318 optlen += tp->rcv_numsacks * TCPOLEN_SACK;
2319 optlen = PAD(optlen);
2320 }
2321 } else {
2322 if (tp->t_flags & TF_REQ_TSTMP)
2323 optlen = TCPOLEN_TSTAMP_APPA;
2324 else
2325 optlen = PAD(TCPOLEN_MAXSEG);
2326 if (tp->t_flags & TF_REQ_SCALE)
2327 optlen += PAD(TCPOLEN_WINDOW);
2328#ifdef TCP_SIGNATURE
2329 if (tp->t_flags & TF_SIGNATURE)
2330 optlen += PAD(TCPOLEN_SIGNATURE);
2331#endif
2332 if (tp->t_flags & TF_SACK_PERMIT)
2333 optlen += PAD(TCPOLEN_SACK_PERMITTED);
2334 }
2335#undef PAD
2336 optlen = min(optlen, TCP_MAXOLEN);
2337 return (tp->t_maxseg - optlen);
2338}
2339
2340#ifdef IPSEC
2341/* compute ESP/AH header size for TCP, including outer IP header. */
2342size_t
2343ipsec_hdrsiz_tcp(struct tcpcb *tp)
2344{
2345 struct inpcb *inp;
2346 struct mbuf *m;
2347 size_t hdrsiz;
2348 struct ip *ip;
2349#ifdef INET6
2350 struct ip6_hdr *ip6;
2351#endif
2352 struct tcphdr *th;
2353
2354 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL) ||
2355 (!key_havesp(IPSEC_DIR_OUTBOUND)))
2356 return (0);
2357 m = m_gethdr(M_NOWAIT, MT_DATA);
2358 if (!m)
2359 return (0);
2360
2361#ifdef INET6
2362 if ((inp->inp_vflag & INP_IPV6) != 0) {
2363 ip6 = mtod(m, struct ip6_hdr *);
2364 th = (struct tcphdr *)(ip6 + 1);
2365 m->m_pkthdr.len = m->m_len =
2366 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
2367 tcpip_fillheaders(inp, ip6, th);
2368 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2369 } else
2370#endif /* INET6 */
2371 {
2372 ip = mtod(m, struct ip *);
2373 th = (struct tcphdr *)(ip + 1);
2374 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
2375 tcpip_fillheaders(inp, ip, th);
2376 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2377 }
2378
2379 m_free(m);
2380 return (hdrsiz);
2381}
2382#endif /* IPSEC */
2383
2384#ifdef TCP_SIGNATURE
2385/*
2386 * Callback function invoked by m_apply() to digest TCP segment data
2387 * contained within an mbuf chain.
2388 */
2389static int
2390tcp_signature_apply(void *fstate, void *data, u_int len)
2391{
2392
2393 MD5Update(fstate, (u_char *)data, len);
2394 return (0);
2395}
2396
2397/*
2398 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
2399 * search with the destination IP address, and a 'magic SPI' to be
2400 * determined by the application. This is hardcoded elsewhere to 1179
2401*/
2402struct secasvar *
2403tcp_get_sav(struct mbuf *m, u_int direction)
2404{
2405 union sockaddr_union dst;
2406 struct secasvar *sav;
2407 struct ip *ip;
2408#ifdef INET6
2409 struct ip6_hdr *ip6;
2410 char ip6buf[INET6_ADDRSTRLEN];
2411#endif
2412
2413 /* Extract the destination from the IP header in the mbuf. */
2414 bzero(&dst, sizeof(union sockaddr_union));
2415 ip = mtod(m, struct ip *);
2416#ifdef INET6
2417 ip6 = NULL; /* Make the compiler happy. */
2418#endif
2419 switch (ip->ip_v) {
2420#ifdef INET
2421 case IPVERSION:
2422 dst.sa.sa_len = sizeof(struct sockaddr_in);
2423 dst.sa.sa_family = AF_INET;
2424 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
2425 ip->ip_src : ip->ip_dst;
2426 break;
2427#endif
2428#ifdef INET6
2429 case (IPV6_VERSION >> 4):
2430 ip6 = mtod(m, struct ip6_hdr *);
2431 dst.sa.sa_len = sizeof(struct sockaddr_in6);
2432 dst.sa.sa_family = AF_INET6;
2433 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
2434 ip6->ip6_src : ip6->ip6_dst;
2435 break;
2436#endif
2437 default:
2438 return (NULL);
2439 /* NOTREACHED */
2440 break;
2441 }
2442
2443 /* Look up an SADB entry which matches the address of the peer. */
2444 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2445 if (sav == NULL) {
2446 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
2447 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
2448#ifdef INET6
2449 (ip->ip_v == (IPV6_VERSION >> 4)) ?
2450 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2451#endif
2452 "(unsupported)"));
2453 }
2454
2455 return (sav);
2456}
2457
2458/*
2459 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2460 *
2461 * Parameters:
2462 * m pointer to head of mbuf chain
2463 * len length of TCP segment data, excluding options
2464 * optlen length of TCP segment options
2465 * buf pointer to storage for computed MD5 digest
2466 * sav pointer to security assosiation
2467 *
2468 * We do this over ip, tcphdr, segment data, and the key in the SADB.
2469 * When called from tcp_input(), we can be sure that th_sum has been
2470 * zeroed out and verified already.
2471 *
2472 * Releases reference to SADB key before return.
2473 *
2474 * Return 0 if successful, otherwise return -1.
2475 *
2476 */
2477int
2478tcp_signature_do_compute(struct mbuf *m, int len, int optlen,
2479 u_char *buf, struct secasvar *sav)
2480{
2481#ifdef INET
2482 struct ippseudo ippseudo;
2483#endif
2484 MD5_CTX ctx;
2485 int doff;
2486 struct ip *ip;
2487#ifdef INET
2488 struct ipovly *ipovly;
2489#endif
2490 struct tcphdr *th;
2491#ifdef INET6
2492 struct ip6_hdr *ip6;
2493 struct in6_addr in6;
2494 uint32_t plen;
2495 uint16_t nhdr;
2496#endif
2497 u_short savecsum;
2498
2499 KASSERT(m != NULL, ("NULL mbuf chain"));
2500 KASSERT(buf != NULL, ("NULL signature pointer"));
2501
2502 /* Extract the destination from the IP header in the mbuf. */
2503 ip = mtod(m, struct ip *);
2504#ifdef INET6
2505 ip6 = NULL; /* Make the compiler happy. */
2506#endif
2507
2508 MD5Init(&ctx);
2509 /*
2510 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2511 *
2512 * XXX The ippseudo header MUST be digested in network byte order,
2513 * or else we'll fail the regression test. Assume all fields we've
2514 * been doing arithmetic on have been in host byte order.
2515 * XXX One cannot depend on ipovly->ih_len here. When called from
2516 * tcp_output(), the underlying ip_len member has not yet been set.
2517 */
2518 switch (ip->ip_v) {
2519#ifdef INET
2520 case IPVERSION:
2521 ipovly = (struct ipovly *)ip;
2522 ippseudo.ippseudo_src = ipovly->ih_src;
2523 ippseudo.ippseudo_dst = ipovly->ih_dst;
2524 ippseudo.ippseudo_pad = 0;
2525 ippseudo.ippseudo_p = IPPROTO_TCP;
2526 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2527 optlen);
2528 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2529
2530 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2531 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2532 break;
2533#endif
2534#ifdef INET6
2535 /*
2536 * RFC 2385, 2.0 Proposal
2537 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2538 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2539 * extended next header value (to form 32 bits), and 32-bit segment
2540 * length.
2541 * Note: Upper-Layer Packet Length comes before Next Header.
2542 */
2543 case (IPV6_VERSION >> 4):
2544 in6 = ip6->ip6_src;
2545 in6_clearscope(&in6);
2546 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2547 in6 = ip6->ip6_dst;
2548 in6_clearscope(&in6);
2549 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2550 plen = htonl(len + sizeof(struct tcphdr) + optlen);
2551 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2552 nhdr = 0;
2553 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2554 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2555 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2556 nhdr = IPPROTO_TCP;
2557 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2558
2559 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2560 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2561 break;
2562#endif
2563 default:
2564 KEY_FREESAV(&sav);
2565 return (-1);
2566 /* NOTREACHED */
2567 break;
2568 }
2569
2570
2571 /*
2572 * Step 2: Update MD5 hash with TCP header, excluding options.
2573 * The TCP checksum must be set to zero.
2574 */
2575 savecsum = th->th_sum;
2576 th->th_sum = 0;
2577 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2578 th->th_sum = savecsum;
2579
2580 /*
2581 * Step 3: Update MD5 hash with TCP segment data.
2582 * Use m_apply() to avoid an early m_pullup().
2583 */
2584 if (len > 0)
2585 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2586
2587 /*
2588 * Step 4: Update MD5 hash with shared secret.
2589 */
2590 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2591 MD5Final(buf, &ctx);
2592
2593 key_sa_recordxfer(sav, m);
2594 KEY_FREESAV(&sav);
2595 return (0);
2596}
2597
2598/*
2599 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2600 *
2601 * Return 0 if successful, otherwise return -1.
2602 */
2603int
2604tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
2605 u_char *buf, u_int direction)
2606{
2607 struct secasvar *sav;
2608
2609 if ((sav = tcp_get_sav(m, direction)) == NULL)
2610 return (-1);
2611
2612 return (tcp_signature_do_compute(m, len, optlen, buf, sav));
2613}
2614
2615/*
2616 * Verify the TCP-MD5 hash of a TCP segment. (RFC2385)
2617 *
2618 * Parameters:
2619 * m pointer to head of mbuf chain
2620 * len length of TCP segment data, excluding options
2621 * optlen length of TCP segment options
2622 * buf pointer to storage for computed MD5 digest
2623 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2624 *
2625 * Return 1 if successful, otherwise return 0.
2626 */
2627int
2628tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen,
2629 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
2630{
2631 char tmpdigest[TCP_SIGLEN];
2632
2633 if (tcp_sig_checksigs == 0)
2634 return (1);
2635 if ((tcpbflag & TF_SIGNATURE) == 0) {
2636 if ((to->to_flags & TOF_SIGNATURE) != 0) {
2637
2638 /*
2639 * If this socket is not expecting signature but
2640 * the segment contains signature just fail.
2641 */
2642 TCPSTAT_INC(tcps_sig_err_sigopt);
2643 TCPSTAT_INC(tcps_sig_rcvbadsig);
2644 return (0);
2645 }
2646
2647 /* Signature is not expected, and not present in segment. */
2648 return (1);
2649 }
2650
2651 /*
2652 * If this socket is expecting signature but the segment does not
2653 * contain any just fail.
2654 */
2655 if ((to->to_flags & TOF_SIGNATURE) == 0) {
2656 TCPSTAT_INC(tcps_sig_err_nosigopt);
2657 TCPSTAT_INC(tcps_sig_rcvbadsig);
2658 return (0);
2659 }
2660 if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0],
2661 IPSEC_DIR_INBOUND) == -1) {
2662 TCPSTAT_INC(tcps_sig_err_buildsig);
2663 TCPSTAT_INC(tcps_sig_rcvbadsig);
2664 return (0);
2665 }
2666
2667 if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) {
2668 TCPSTAT_INC(tcps_sig_rcvbadsig);
2669 return (0);
2670 }
2671 TCPSTAT_INC(tcps_sig_rcvgoodsig);
2672 return (1);
2673}
2674#endif /* TCP_SIGNATURE */
2675
2676static int
2677sysctl_drop(SYSCTL_HANDLER_ARGS)
2678{
2679 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2680 struct sockaddr_storage addrs[2];
2681 struct inpcb *inp;
2682 struct tcpcb *tp;
2683 struct tcptw *tw;
2684 struct sockaddr_in *fin, *lin;
2685#ifdef INET6
2686 struct sockaddr_in6 *fin6, *lin6;
2687#endif
2688 int error;
2689
2690 inp = NULL;
2691 fin = lin = NULL;
2692#ifdef INET6
2693 fin6 = lin6 = NULL;
2694#endif
2695 error = 0;
2696
2697 if (req->oldptr != NULL || req->oldlen != 0)
2698 return (EINVAL);
2699 if (req->newptr == NULL)
2700 return (EPERM);
2701 if (req->newlen < sizeof(addrs))
2702 return (ENOMEM);
2703 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2704 if (error)
2705 return (error);
2706
2707 switch (addrs[0].ss_family) {
2708#ifdef INET6
2709 case AF_INET6:
2710 fin6 = (struct sockaddr_in6 *)&addrs[0];
2711 lin6 = (struct sockaddr_in6 *)&addrs[1];
2712 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2713 lin6->sin6_len != sizeof(struct sockaddr_in6))
2714 return (EINVAL);
2715 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2716 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2717 return (EINVAL);
2718 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2719 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2720 fin = (struct sockaddr_in *)&addrs[0];
2721 lin = (struct sockaddr_in *)&addrs[1];
2722 break;
2723 }
2724 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2725 if (error)
2726 return (error);
2727 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2728 if (error)
2729 return (error);
2730 break;
2731#endif
2732#ifdef INET
2733 case AF_INET:
2734 fin = (struct sockaddr_in *)&addrs[0];
2735 lin = (struct sockaddr_in *)&addrs[1];
2736 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2737 lin->sin_len != sizeof(struct sockaddr_in))
2738 return (EINVAL);
2739 break;
2740#endif
2741 default:
2742 return (EINVAL);
2743 }
2744 INP_INFO_RLOCK(&V_tcbinfo);
2745 switch (addrs[0].ss_family) {
2746#ifdef INET6
2747 case AF_INET6:
2748 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2749 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2750 INPLOOKUP_WLOCKPCB, NULL);
2751 break;
2752#endif
2753#ifdef INET
2754 case AF_INET:
2755 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2756 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2757 break;
2758#endif
2759 }
2760 if (inp != NULL) {
2761 if (inp->inp_flags & INP_TIMEWAIT) {
2762 /*
2763 * XXXRW: There currently exists a state where an
2764 * inpcb is present, but its timewait state has been
2765 * discarded. For now, don't allow dropping of this
2766 * type of inpcb.
2767 */
2768 tw = intotw(inp);
2769 if (tw != NULL)
2770 tcp_twclose(tw, 0);
2771 else
2772 INP_WUNLOCK(inp);
2773 } else if (!(inp->inp_flags & INP_DROPPED) &&
2774 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2775 tp = intotcpcb(inp);
2776 tp = tcp_drop(tp, ECONNABORTED);
2777 if (tp != NULL)
2778 INP_WUNLOCK(inp);
2779 } else
2780 INP_WUNLOCK(inp);
2781 } else
2782 error = ESRCH;
2783 INP_INFO_RUNLOCK(&V_tcbinfo);
2784 return (error);
2785}
2786
2787SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2788 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
2789 0, sysctl_drop, "", "Drop TCP connection");
2790
2791/*
2792 * Generate a standardized TCP log line for use throughout the
2793 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2794 * allow use in the interrupt context.
2795 *
2796 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2797 * NB: The function may return NULL if memory allocation failed.
2798 *
2799 * Due to header inclusion and ordering limitations the struct ip
2800 * and ip6_hdr pointers have to be passed as void pointers.
2801 */
2802char *
2803tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2804 const void *ip6hdr)
2805{
2806
2807 /* Is logging enabled? */
2808 if (tcp_log_in_vain == 0)
2809 return (NULL);
2810
2811 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2812}
2813
2814char *
2815tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2816 const void *ip6hdr)
2817{
2818
2819 /* Is logging enabled? */
2820 if (tcp_log_debug == 0)
2821 return (NULL);
2822
2823 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2824}
2825
2826static char *
2827tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2828 const void *ip6hdr)
2829{
2830 char *s, *sp;
2831 size_t size;
2832 struct ip *ip;
2833#ifdef INET6
2834 const struct ip6_hdr *ip6;
2835
2836 ip6 = (const struct ip6_hdr *)ip6hdr;
2837#endif /* INET6 */
2838 ip = (struct ip *)ip4hdr;
2839
2840 /*
2841 * The log line looks like this:
2842 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2843 */
2844 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2845 sizeof(PRINT_TH_FLAGS) + 1 +
2846#ifdef INET6
2847 2 * INET6_ADDRSTRLEN;
2848#else
2849 2 * INET_ADDRSTRLEN;
2850#endif /* INET6 */
2851
2852 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2853 if (s == NULL)
2854 return (NULL);
2855
2856 strcat(s, "TCP: [");
2857 sp = s + strlen(s);
2858
2859 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2860 inet_ntoa_r(inc->inc_faddr, sp);
2861 sp = s + strlen(s);
2862 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2863 sp = s + strlen(s);
2864 inet_ntoa_r(inc->inc_laddr, sp);
2865 sp = s + strlen(s);
2866 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2867#ifdef INET6
2868 } else if (inc) {
2869 ip6_sprintf(sp, &inc->inc6_faddr);
2870 sp = s + strlen(s);
2871 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2872 sp = s + strlen(s);
2873 ip6_sprintf(sp, &inc->inc6_laddr);
2874 sp = s + strlen(s);
2875 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2876 } else if (ip6 && th) {
2877 ip6_sprintf(sp, &ip6->ip6_src);
2878 sp = s + strlen(s);
2879 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2880 sp = s + strlen(s);
2881 ip6_sprintf(sp, &ip6->ip6_dst);
2882 sp = s + strlen(s);
2883 sprintf(sp, "]:%i", ntohs(th->th_dport));
2884#endif /* INET6 */
2885#ifdef INET
2886 } else if (ip && th) {
2887 inet_ntoa_r(ip->ip_src, sp);
2888 sp = s + strlen(s);
2889 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2890 sp = s + strlen(s);
2891 inet_ntoa_r(ip->ip_dst, sp);
2892 sp = s + strlen(s);
2893 sprintf(sp, "]:%i", ntohs(th->th_dport));
2894#endif /* INET */
2895 } else {
2896 free(s, M_TCPLOG);
2897 return (NULL);
2898 }
2899 sp = s + strlen(s);
2900 if (th)
2901 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2902 if (*(s + size - 1) != '\0')
2903 panic("%s: string too long", __func__);
2904 return (s);
2905}
2906
2907/*
2908 * A subroutine which makes it easy to track TCP state changes with DTrace.
2909 * This function shouldn't be called for t_state initializations that don't
2910 * correspond to actual TCP state transitions.
2911 */
2912void
2913tcp_state_change(struct tcpcb *tp, int newstate)
2914{
2915#if defined(KDTRACE_HOOKS)
2916 int pstate = tp->t_state;
2917#endif
2918
2919 tp->t_state = newstate;
2920 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
2921}
| 97#ifdef INET6
98#include <netinet6/tcp6_var.h>
99#endif
100#include <netinet/tcpip.h>
101#ifdef TCPPCAP
102#include <netinet/tcp_pcap.h>
103#endif
104#ifdef TCPDEBUG
105#include <netinet/tcp_debug.h>
106#endif
107#ifdef INET6
108#include <netinet6/ip6protosw.h>
109#endif
110#ifdef TCP_OFFLOAD
111#include <netinet/tcp_offload.h>
112#endif
113
114#ifdef IPSEC
115#include <netipsec/ipsec.h>
116#include <netipsec/xform.h>
117#ifdef INET6
118#include <netipsec/ipsec6.h>
119#endif
120#include <netipsec/key.h>
121#include <sys/syslog.h>
122#endif /*IPSEC*/
123
124#include <machine/in_cksum.h>
125#include <sys/md5.h>
126
127#include <security/mac/mac_framework.h>
128
129VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
130#ifdef INET6
131VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
132#endif
133
134struct rwlock tcp_function_lock;
135
136static int
137sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
138{
139 int error, new;
140
141 new = V_tcp_mssdflt;
142 error = sysctl_handle_int(oidp, &new, 0, req);
143 if (error == 0 && req->newptr) {
144 if (new < TCP_MINMSS)
145 error = EINVAL;
146 else
147 V_tcp_mssdflt = new;
148 }
149 return (error);
150}
151
152SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
153 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
154 &sysctl_net_inet_tcp_mss_check, "I",
155 "Default TCP Maximum Segment Size");
156
157#ifdef INET6
158static int
159sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
160{
161 int error, new;
162
163 new = V_tcp_v6mssdflt;
164 error = sysctl_handle_int(oidp, &new, 0, req);
165 if (error == 0 && req->newptr) {
166 if (new < TCP_MINMSS)
167 error = EINVAL;
168 else
169 V_tcp_v6mssdflt = new;
170 }
171 return (error);
172}
173
174SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
175 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
176 &sysctl_net_inet_tcp_mss_v6_check, "I",
177 "Default TCP Maximum Segment Size for IPv6");
178#endif /* INET6 */
179
180/*
181 * Minimum MSS we accept and use. This prevents DoS attacks where
182 * we are forced to a ridiculous low MSS like 20 and send hundreds
183 * of packets instead of one. The effect scales with the available
184 * bandwidth and quickly saturates the CPU and network interface
185 * with packet generation and sending. Set to zero to disable MINMSS
186 * checking. This setting prevents us from sending too small packets.
187 */
188VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
189SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
190 &VNET_NAME(tcp_minmss), 0,
191 "Minimum TCP Maximum Segment Size");
192
193VNET_DEFINE(int, tcp_do_rfc1323) = 1;
194SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
195 &VNET_NAME(tcp_do_rfc1323), 0,
196 "Enable rfc1323 (high performance TCP) extensions");
197
198static int tcp_log_debug = 0;
199SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
200 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
201
202static int tcp_tcbhashsize;
203SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
204 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
205
206static int do_tcpdrain = 1;
207SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
208 "Enable tcp_drain routine for extra help when low on mbufs");
209
210SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
211 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
212
213static VNET_DEFINE(int, icmp_may_rst) = 1;
214#define V_icmp_may_rst VNET(icmp_may_rst)
215SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
216 &VNET_NAME(icmp_may_rst), 0,
217 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
218
219static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
220#define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
221SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
222 &VNET_NAME(tcp_isn_reseed_interval), 0,
223 "Seconds between reseeding of ISN secret");
224
225static int tcp_soreceive_stream;
226SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
227 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
228
229#ifdef TCP_SIGNATURE
230static int tcp_sig_checksigs = 1;
231SYSCTL_INT(_net_inet_tcp, OID_AUTO, signature_verify_input, CTLFLAG_RW,
232 &tcp_sig_checksigs, 0, "Verify RFC2385 digests on inbound traffic");
233#endif
234
235VNET_DEFINE(uma_zone_t, sack_hole_zone);
236#define V_sack_hole_zone VNET(sack_hole_zone)
237
238VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
239
240static struct inpcb *tcp_notify(struct inpcb *, int);
241static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
242static void tcp_mtudisc(struct inpcb *, int);
243static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
244 void *ip4hdr, const void *ip6hdr);
245static void tcp_timer_discard(struct tcpcb *, uint32_t);
246
247
248static struct tcp_function_block tcp_def_funcblk = {
249 "default",
250 tcp_output,
251 tcp_do_segment,
252 tcp_default_ctloutput,
253 NULL,
254 NULL,
255 NULL,
256 NULL,
257 NULL,
258 NULL,
259 NULL,
260 0,
261 0
262};
263
264struct tcp_funchead t_functions;
265static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
266
267static struct tcp_function_block *
268find_tcp_functions_locked(struct tcp_function_set *fs)
269{
270 struct tcp_function *f;
271 struct tcp_function_block *blk=NULL;
272
273 TAILQ_FOREACH(f, &t_functions, tf_next) {
274 if (strcmp(f->tf_fb->tfb_tcp_block_name, fs->function_set_name) == 0) {
275 blk = f->tf_fb;
276 break;
277 }
278 }
279 return(blk);
280}
281
282static struct tcp_function_block *
283find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
284{
285 struct tcp_function_block *rblk=NULL;
286 struct tcp_function *f;
287
288 TAILQ_FOREACH(f, &t_functions, tf_next) {
289 if (f->tf_fb == blk) {
290 rblk = blk;
291 if (s) {
292 *s = f;
293 }
294 break;
295 }
296 }
297 return (rblk);
298}
299
300struct tcp_function_block *
301find_and_ref_tcp_functions(struct tcp_function_set *fs)
302{
303 struct tcp_function_block *blk;
304
305 rw_rlock(&tcp_function_lock);
306 blk = find_tcp_functions_locked(fs);
307 if (blk)
308 refcount_acquire(&blk->tfb_refcnt);
309 rw_runlock(&tcp_function_lock);
310 return(blk);
311}
312
313struct tcp_function_block *
314find_and_ref_tcp_fb(struct tcp_function_block *blk)
315{
316 struct tcp_function_block *rblk;
317
318 rw_rlock(&tcp_function_lock);
319 rblk = find_tcp_fb_locked(blk, NULL);
320 if (rblk)
321 refcount_acquire(&rblk->tfb_refcnt);
322 rw_runlock(&tcp_function_lock);
323 return(rblk);
324}
325
326
327static int
328sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
329{
330 int error=ENOENT;
331 struct tcp_function_set fs;
332 struct tcp_function_block *blk;
333
334 memset(&fs, 0, sizeof(fs));
335 rw_rlock(&tcp_function_lock);
336 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
337 if (blk) {
338 /* Found him */
339 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
340 fs.pcbcnt = blk->tfb_refcnt;
341 }
342 rw_runlock(&tcp_function_lock);
343 error = sysctl_handle_string(oidp, fs.function_set_name,
344 sizeof(fs.function_set_name), req);
345
346 /* Check for error or no change */
347 if (error != 0 || req->newptr == NULL)
348 return(error);
349
350 rw_wlock(&tcp_function_lock);
351 blk = find_tcp_functions_locked(&fs);
352 if ((blk == NULL) ||
353 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
354 error = ENOENT;
355 goto done;
356 }
357 tcp_func_set_ptr = blk;
358done:
359 rw_wunlock(&tcp_function_lock);
360 return (error);
361}
362
363SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
364 CTLTYPE_STRING | CTLFLAG_RW,
365 NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
366 "Set/get the default TCP functions");
367
368static int
369sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
370{
371 int error, cnt, linesz;
372 struct tcp_function *f;
373 char *buffer, *cp;
374 size_t bufsz, outsz;
375
376 cnt = 0;
377 rw_rlock(&tcp_function_lock);
378 TAILQ_FOREACH(f, &t_functions, tf_next) {
379 cnt++;
380 }
381 rw_runlock(&tcp_function_lock);
382
383 bufsz = (cnt+2) * (TCP_FUNCTION_NAME_LEN_MAX + 12) + 1;
384 buffer = malloc(bufsz, M_TEMP, M_WAITOK);
385
386 error = 0;
387 cp = buffer;
388
389 linesz = snprintf(cp, bufsz, "\n%-32s%c %s\n", "Stack", 'D', "PCB count");
390 cp += linesz;
391 bufsz -= linesz;
392 outsz = linesz;
393
394 rw_rlock(&tcp_function_lock);
395 TAILQ_FOREACH(f, &t_functions, tf_next) {
396 linesz = snprintf(cp, bufsz, "%-32s%c %u\n",
397 f->tf_fb->tfb_tcp_block_name,
398 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
399 f->tf_fb->tfb_refcnt);
400 if (linesz >= bufsz) {
401 error = EOVERFLOW;
402 break;
403 }
404 cp += linesz;
405 bufsz -= linesz;
406 outsz += linesz;
407 }
408 rw_runlock(&tcp_function_lock);
409 if (error == 0)
410 error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
411 free(buffer, M_TEMP);
412 return (error);
413}
414
415SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
416 CTLTYPE_STRING|CTLFLAG_RD,
417 NULL, 0, sysctl_net_inet_list_available, "A",
418 "list available TCP Function sets");
419
420/*
421 * Target size of TCP PCB hash tables. Must be a power of two.
422 *
423 * Note that this can be overridden by the kernel environment
424 * variable net.inet.tcp.tcbhashsize
425 */
426#ifndef TCBHASHSIZE
427#define TCBHASHSIZE 0
428#endif
429
430/*
431 * XXX
432 * Callouts should be moved into struct tcp directly. They are currently
433 * separate because the tcpcb structure is exported to userland for sysctl
434 * parsing purposes, which do not know about callouts.
435 */
436struct tcpcb_mem {
437 struct tcpcb tcb;
438 struct tcp_timer tt;
439 struct cc_var ccv;
440 struct osd osd;
441};
442
443static VNET_DEFINE(uma_zone_t, tcpcb_zone);
444#define V_tcpcb_zone VNET(tcpcb_zone)
445
446MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
447MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
448
449static struct mtx isn_mtx;
450
451#define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
452#define ISN_LOCK() mtx_lock(&isn_mtx)
453#define ISN_UNLOCK() mtx_unlock(&isn_mtx)
454
455/*
456 * TCP initialization.
457 */
458static void
459tcp_zone_change(void *tag)
460{
461
462 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
463 uma_zone_set_max(V_tcpcb_zone, maxsockets);
464 tcp_tw_zone_change();
465}
466
467static int
468tcp_inpcb_init(void *mem, int size, int flags)
469{
470 struct inpcb *inp = mem;
471
472 INP_LOCK_INIT(inp, "inp", "tcpinp");
473 return (0);
474}
475
476/*
477 * Take a value and get the next power of 2 that doesn't overflow.
478 * Used to size the tcp_inpcb hash buckets.
479 */
480static int
481maketcp_hashsize(int size)
482{
483 int hashsize;
484
485 /*
486 * auto tune.
487 * get the next power of 2 higher than maxsockets.
488 */
489 hashsize = 1 << fls(size);
490 /* catch overflow, and just go one power of 2 smaller */
491 if (hashsize < size) {
492 hashsize = 1 << (fls(size) - 1);
493 }
494 return (hashsize);
495}
496
497int
498register_tcp_functions(struct tcp_function_block *blk, int wait)
499{
500 struct tcp_function_block *lblk;
501 struct tcp_function *n;
502 struct tcp_function_set fs;
503
504 if ((blk->tfb_tcp_output == NULL) ||
505 (blk->tfb_tcp_do_segment == NULL) ||
506 (blk->tfb_tcp_ctloutput == NULL) ||
507 (strlen(blk->tfb_tcp_block_name) == 0)) {
508 /*
509 * These functions are required and you
510 * need a name.
511 */
512 return (EINVAL);
513 }
514 if (blk->tfb_tcp_timer_stop_all ||
515 blk->tfb_tcp_timers_left ||
516 blk->tfb_tcp_timer_activate ||
517 blk->tfb_tcp_timer_active ||
518 blk->tfb_tcp_timer_stop) {
519 /*
520 * If you define one timer function you
521 * must have them all.
522 */
523 if ((blk->tfb_tcp_timer_stop_all == NULL) ||
524 (blk->tfb_tcp_timers_left == NULL) ||
525 (blk->tfb_tcp_timer_activate == NULL) ||
526 (blk->tfb_tcp_timer_active == NULL) ||
527 (blk->tfb_tcp_timer_stop == NULL)) {
528 return (EINVAL);
529 }
530 }
531 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
532 if (n == NULL) {
533 return (ENOMEM);
534 }
535 n->tf_fb = blk;
536 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
537 rw_wlock(&tcp_function_lock);
538 lblk = find_tcp_functions_locked(&fs);
539 if (lblk) {
540 /* Duplicate name space not allowed */
541 rw_wunlock(&tcp_function_lock);
542 free(n, M_TCPFUNCTIONS);
543 return (EALREADY);
544 }
545 refcount_init(&blk->tfb_refcnt, 0);
546 blk->tfb_flags = 0;
547 TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
548 rw_wunlock(&tcp_function_lock);
549 return(0);
550}
551
552int
553deregister_tcp_functions(struct tcp_function_block *blk)
554{
555 struct tcp_function_block *lblk;
556 struct tcp_function *f;
557 int error=ENOENT;
558
559 if (strcmp(blk->tfb_tcp_block_name, "default") == 0) {
560 /* You can't un-register the default */
561 return (EPERM);
562 }
563 rw_wlock(&tcp_function_lock);
564 if (blk == tcp_func_set_ptr) {
565 /* You can't free the current default */
566 rw_wunlock(&tcp_function_lock);
567 return (EBUSY);
568 }
569 if (blk->tfb_refcnt) {
570 /* Still tcb attached, mark it. */
571 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
572 rw_wunlock(&tcp_function_lock);
573 return (EBUSY);
574 }
575 lblk = find_tcp_fb_locked(blk, &f);
576 if (lblk) {
577 /* Found */
578 TAILQ_REMOVE(&t_functions, f, tf_next);
579 f->tf_fb = NULL;
580 free(f, M_TCPFUNCTIONS);
581 error = 0;
582 }
583 rw_wunlock(&tcp_function_lock);
584 return (error);
585}
586
587void
588tcp_init(void)
589{
590 const char *tcbhash_tuneable;
591 int hashsize;
592
593 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
594
595 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
596 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
597 printf("%s: WARNING: unable to register helper hook\n", __func__);
598 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
599 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
600 printf("%s: WARNING: unable to register helper hook\n", __func__);
601 hashsize = TCBHASHSIZE;
602 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
603 if (hashsize == 0) {
604 /*
605 * Auto tune the hash size based on maxsockets.
606 * A perfect hash would have a 1:1 mapping
607 * (hashsize = maxsockets) however it's been
608 * suggested that O(2) average is better.
609 */
610 hashsize = maketcp_hashsize(maxsockets / 4);
611 /*
612 * Our historical default is 512,
613 * do not autotune lower than this.
614 */
615 if (hashsize < 512)
616 hashsize = 512;
617 if (bootverbose && IS_DEFAULT_VNET(curvnet))
618 printf("%s: %s auto tuned to %d\n", __func__,
619 tcbhash_tuneable, hashsize);
620 }
621 /*
622 * We require a hashsize to be a power of two.
623 * Previously if it was not a power of two we would just reset it
624 * back to 512, which could be a nasty surprise if you did not notice
625 * the error message.
626 * Instead what we do is clip it to the closest power of two lower
627 * than the specified hash value.
628 */
629 if (!powerof2(hashsize)) {
630 int oldhashsize = hashsize;
631
632 hashsize = maketcp_hashsize(hashsize);
633 /* prevent absurdly low value */
634 if (hashsize < 16)
635 hashsize = 16;
636 printf("%s: WARNING: TCB hash size not a power of 2, "
637 "clipped from %d to %d.\n", __func__, oldhashsize,
638 hashsize);
639 }
640 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
641 "tcp_inpcb", tcp_inpcb_init, NULL, UMA_ZONE_NOFREE,
642 IPI_HASHFIELDS_4TUPLE);
643
644 /*
645 * These have to be type stable for the benefit of the timers.
646 */
647 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
648 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
649 uma_zone_set_max(V_tcpcb_zone, maxsockets);
650 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
651
652 tcp_tw_init();
653 syncache_init();
654 tcp_hc_init();
655
656 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
657 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
658 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
659
660 /* Skip initialization of globals for non-default instances. */
661 if (!IS_DEFAULT_VNET(curvnet))
662 return;
663
664 tcp_reass_global_init();
665
666 /* XXX virtualize those bellow? */
667 tcp_delacktime = TCPTV_DELACK;
668 tcp_keepinit = TCPTV_KEEP_INIT;
669 tcp_keepidle = TCPTV_KEEP_IDLE;
670 tcp_keepintvl = TCPTV_KEEPINTVL;
671 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
672 tcp_msl = TCPTV_MSL;
673 tcp_rexmit_min = TCPTV_MIN;
674 if (tcp_rexmit_min < 1)
675 tcp_rexmit_min = 1;
676 tcp_rexmit_slop = TCPTV_CPU_VAR;
677 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
678 tcp_tcbhashsize = hashsize;
679 /* Setup the tcp function block list */
680 TAILQ_INIT(&t_functions);
681 rw_init_flags(&tcp_function_lock, "tcp_func_lock" , 0);
682 register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
683
684 if (tcp_soreceive_stream) {
685#ifdef INET
686 tcp_usrreqs.pru_soreceive = soreceive_stream;
687#endif
688#ifdef INET6
689 tcp6_usrreqs.pru_soreceive = soreceive_stream;
690#endif /* INET6 */
691 }
692
693#ifdef INET6
694#define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
695#else /* INET6 */
696#define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
697#endif /* INET6 */
698 if (max_protohdr < TCP_MINPROTOHDR)
699 max_protohdr = TCP_MINPROTOHDR;
700 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
701 panic("tcp_init");
702#undef TCP_MINPROTOHDR
703
704 ISN_LOCK_INIT();
705 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
706 SHUTDOWN_PRI_DEFAULT);
707 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
708 EVENTHANDLER_PRI_ANY);
709#ifdef TCPPCAP
710 tcp_pcap_init();
711#endif
712
713#ifdef TCP_RFC7413
714 tcp_fastopen_init();
715#endif
716}
717
718#ifdef VIMAGE
719void
720tcp_destroy(void)
721{
722 int error;
723
724#ifdef TCP_RFC7413
725 tcp_fastopen_destroy();
726#endif
727 tcp_hc_destroy();
728 syncache_destroy();
729 tcp_tw_destroy();
730 in_pcbinfo_destroy(&V_tcbinfo);
731 uma_zdestroy(V_sack_hole_zone);
732 uma_zdestroy(V_tcpcb_zone);
733
734 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
735 if (error != 0) {
736 printf("%s: WARNING: unable to deregister helper hook "
737 "type=%d, id=%d: error %d returned\n", __func__,
738 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
739 }
740 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
741 if (error != 0) {
742 printf("%s: WARNING: unable to deregister helper hook "
743 "type=%d, id=%d: error %d returned\n", __func__,
744 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
745 }
746}
747#endif
748
749void
750tcp_fini(void *xtp)
751{
752
753}
754
755/*
756 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
757 * tcp_template used to store this data in mbufs, but we now recopy it out
758 * of the tcpcb each time to conserve mbufs.
759 */
760void
761tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
762{
763 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
764
765 INP_WLOCK_ASSERT(inp);
766
767#ifdef INET6
768 if ((inp->inp_vflag & INP_IPV6) != 0) {
769 struct ip6_hdr *ip6;
770
771 ip6 = (struct ip6_hdr *)ip_ptr;
772 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
773 (inp->inp_flow & IPV6_FLOWINFO_MASK);
774 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
775 (IPV6_VERSION & IPV6_VERSION_MASK);
776 ip6->ip6_nxt = IPPROTO_TCP;
777 ip6->ip6_plen = htons(sizeof(struct tcphdr));
778 ip6->ip6_src = inp->in6p_laddr;
779 ip6->ip6_dst = inp->in6p_faddr;
780 }
781#endif /* INET6 */
782#if defined(INET6) && defined(INET)
783 else
784#endif
785#ifdef INET
786 {
787 struct ip *ip;
788
789 ip = (struct ip *)ip_ptr;
790 ip->ip_v = IPVERSION;
791 ip->ip_hl = 5;
792 ip->ip_tos = inp->inp_ip_tos;
793 ip->ip_len = 0;
794 ip->ip_id = 0;
795 ip->ip_off = 0;
796 ip->ip_ttl = inp->inp_ip_ttl;
797 ip->ip_sum = 0;
798 ip->ip_p = IPPROTO_TCP;
799 ip->ip_src = inp->inp_laddr;
800 ip->ip_dst = inp->inp_faddr;
801 }
802#endif /* INET */
803 th->th_sport = inp->inp_lport;
804 th->th_dport = inp->inp_fport;
805 th->th_seq = 0;
806 th->th_ack = 0;
807 th->th_x2 = 0;
808 th->th_off = 5;
809 th->th_flags = 0;
810 th->th_win = 0;
811 th->th_urp = 0;
812 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
813}
814
815/*
816 * Create template to be used to send tcp packets on a connection.
817 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
818 * use for this function is in keepalives, which use tcp_respond.
819 */
820struct tcptemp *
821tcpip_maketemplate(struct inpcb *inp)
822{
823 struct tcptemp *t;
824
825 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
826 if (t == NULL)
827 return (NULL);
828 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
829 return (t);
830}
831
832/*
833 * Send a single message to the TCP at address specified by
834 * the given TCP/IP header. If m == NULL, then we make a copy
835 * of the tcpiphdr at th and send directly to the addressed host.
836 * This is used to force keep alive messages out using the TCP
837 * template for a connection. If flags are given then we send
838 * a message back to the TCP which originated the segment th,
839 * and discard the mbuf containing it and any other attached mbufs.
840 *
841 * In any case the ack and sequence number of the transmitted
842 * segment are as specified by the parameters.
843 *
844 * NOTE: If m != NULL, then th must point to *inside* the mbuf.
845 */
846void
847tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
848 tcp_seq ack, tcp_seq seq, int flags)
849{
850 int tlen;
851 int win = 0;
852 struct ip *ip;
853 struct tcphdr *nth;
854#ifdef INET6
855 struct ip6_hdr *ip6;
856 int isipv6;
857#endif /* INET6 */
858 int ipflags = 0;
859 struct inpcb *inp;
860
861 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
862
863#ifdef INET6
864 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
865 ip6 = ipgen;
866#endif /* INET6 */
867 ip = ipgen;
868
869 if (tp != NULL) {
870 inp = tp->t_inpcb;
871 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
872 INP_WLOCK_ASSERT(inp);
873 } else
874 inp = NULL;
875
876 if (tp != NULL) {
877 if (!(flags & TH_RST)) {
878 win = sbspace(&inp->inp_socket->so_rcv);
879 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
880 win = (long)TCP_MAXWIN << tp->rcv_scale;
881 }
882 }
883 if (m == NULL) {
884 m = m_gethdr(M_NOWAIT, MT_DATA);
885 if (m == NULL)
886 return;
887 tlen = 0;
888 m->m_data += max_linkhdr;
889#ifdef INET6
890 if (isipv6) {
891 bcopy((caddr_t)ip6, mtod(m, caddr_t),
892 sizeof(struct ip6_hdr));
893 ip6 = mtod(m, struct ip6_hdr *);
894 nth = (struct tcphdr *)(ip6 + 1);
895 } else
896#endif /* INET6 */
897 {
898 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
899 ip = mtod(m, struct ip *);
900 nth = (struct tcphdr *)(ip + 1);
901 }
902 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
903 flags = TH_ACK;
904 } else {
905 /*
906 * reuse the mbuf.
907 * XXX MRT We inherrit the FIB, which is lucky.
908 */
909 m_freem(m->m_next);
910 m->m_next = NULL;
911 m->m_data = (caddr_t)ipgen;
912 /* m_len is set later */
913 tlen = 0;
914#define xchg(a,b,type) { type t; t=a; a=b; b=t; }
915#ifdef INET6
916 if (isipv6) {
917 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
918 nth = (struct tcphdr *)(ip6 + 1);
919 } else
920#endif /* INET6 */
921 {
922 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
923 nth = (struct tcphdr *)(ip + 1);
924 }
925 if (th != nth) {
926 /*
927 * this is usually a case when an extension header
928 * exists between the IPv6 header and the
929 * TCP header.
930 */
931 nth->th_sport = th->th_sport;
932 nth->th_dport = th->th_dport;
933 }
934 xchg(nth->th_dport, nth->th_sport, uint16_t);
935#undef xchg
936 }
937#ifdef INET6
938 if (isipv6) {
939 ip6->ip6_flow = 0;
940 ip6->ip6_vfc = IPV6_VERSION;
941 ip6->ip6_nxt = IPPROTO_TCP;
942 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
943 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
944 }
945#endif
946#if defined(INET) && defined(INET6)
947 else
948#endif
949#ifdef INET
950 {
951 tlen += sizeof (struct tcpiphdr);
952 ip->ip_len = htons(tlen);
953 ip->ip_ttl = V_ip_defttl;
954 if (V_path_mtu_discovery)
955 ip->ip_off |= htons(IP_DF);
956 }
957#endif
958 m->m_len = tlen;
959 m->m_pkthdr.len = tlen;
960 m->m_pkthdr.rcvif = NULL;
961#ifdef MAC
962 if (inp != NULL) {
963 /*
964 * Packet is associated with a socket, so allow the
965 * label of the response to reflect the socket label.
966 */
967 INP_WLOCK_ASSERT(inp);
968 mac_inpcb_create_mbuf(inp, m);
969 } else {
970 /*
971 * Packet is not associated with a socket, so possibly
972 * update the label in place.
973 */
974 mac_netinet_tcp_reply(m);
975 }
976#endif
977 nth->th_seq = htonl(seq);
978 nth->th_ack = htonl(ack);
979 nth->th_x2 = 0;
980 nth->th_off = sizeof (struct tcphdr) >> 2;
981 nth->th_flags = flags;
982 if (tp != NULL)
983 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
984 else
985 nth->th_win = htons((u_short)win);
986 nth->th_urp = 0;
987
988 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
989#ifdef INET6
990 if (isipv6) {
991 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
992 nth->th_sum = in6_cksum_pseudo(ip6,
993 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
994 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
995 NULL, NULL);
996 }
997#endif /* INET6 */
998#if defined(INET6) && defined(INET)
999 else
1000#endif
1001#ifdef INET
1002 {
1003 m->m_pkthdr.csum_flags = CSUM_TCP;
1004 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1005 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1006 }
1007#endif /* INET */
1008#ifdef TCPDEBUG
1009 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1010 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1011#endif
1012 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *));
1013 if (flags & TH_RST)
1014 TCP_PROBE5(accept__refused, NULL, NULL, mtod(m, const char *),
1015 tp, nth);
1016
1017 TCP_PROBE5(send, NULL, tp, mtod(m, const char *), tp, nth);
1018#ifdef INET6
1019 if (isipv6)
1020 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
1021#endif /* INET6 */
1022#if defined(INET) && defined(INET6)
1023 else
1024#endif
1025#ifdef INET
1026 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
1027#endif
1028}
1029
1030/*
1031 * Create a new TCP control block, making an
1032 * empty reassembly queue and hooking it to the argument
1033 * protocol control block. The `inp' parameter must have
1034 * come from the zone allocator set up in tcp_init().
1035 */
1036struct tcpcb *
1037tcp_newtcpcb(struct inpcb *inp)
1038{
1039 struct tcpcb_mem *tm;
1040 struct tcpcb *tp;
1041#ifdef INET6
1042 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1043#endif /* INET6 */
1044
1045 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1046 if (tm == NULL)
1047 return (NULL);
1048 tp = &tm->tcb;
1049
1050 /* Initialise cc_var struct for this tcpcb. */
1051 tp->ccv = &tm->ccv;
1052 tp->ccv->type = IPPROTO_TCP;
1053 tp->ccv->ccvc.tcp = tp;
1054 rw_rlock(&tcp_function_lock);
1055 tp->t_fb = tcp_func_set_ptr;
1056 refcount_acquire(&tp->t_fb->tfb_refcnt);
1057 rw_runlock(&tcp_function_lock);
1058 if (tp->t_fb->tfb_tcp_fb_init) {
1059 (*tp->t_fb->tfb_tcp_fb_init)(tp);
1060 }
1061 /*
1062 * Use the current system default CC algorithm.
1063 */
1064 CC_LIST_RLOCK();
1065 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1066 CC_ALGO(tp) = CC_DEFAULT();
1067 CC_LIST_RUNLOCK();
1068
1069 if (CC_ALGO(tp)->cb_init != NULL)
1070 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1071 if (tp->t_fb->tfb_tcp_fb_fini)
1072 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1073 refcount_release(&tp->t_fb->tfb_refcnt);
1074 uma_zfree(V_tcpcb_zone, tm);
1075 return (NULL);
1076 }
1077
1078 tp->osd = &tm->osd;
1079 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1080 if (tp->t_fb->tfb_tcp_fb_fini)
1081 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1082 refcount_release(&tp->t_fb->tfb_refcnt);
1083 uma_zfree(V_tcpcb_zone, tm);
1084 return (NULL);
1085 }
1086
1087#ifdef VIMAGE
1088 tp->t_vnet = inp->inp_vnet;
1089#endif
1090 tp->t_timers = &tm->tt;
1091 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
1092 tp->t_maxseg =
1093#ifdef INET6
1094 isipv6 ? V_tcp_v6mssdflt :
1095#endif /* INET6 */
1096 V_tcp_mssdflt;
1097
1098 /* Set up our timeouts. */
1099 callout_init(&tp->t_timers->tt_rexmt, 1);
1100 callout_init(&tp->t_timers->tt_persist, 1);
1101 callout_init(&tp->t_timers->tt_keep, 1);
1102 callout_init(&tp->t_timers->tt_2msl, 1);
1103 callout_init(&tp->t_timers->tt_delack, 1);
1104
1105 if (V_tcp_do_rfc1323)
1106 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1107 if (V_tcp_do_sack)
1108 tp->t_flags |= TF_SACK_PERMIT;
1109 TAILQ_INIT(&tp->snd_holes);
1110 /*
1111 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1112 * is called.
1113 */
1114 in_pcbref(inp); /* Reference for tcpcb */
1115 tp->t_inpcb = inp;
1116
1117 /*
1118 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1119 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
1120 * reasonable initial retransmit time.
1121 */
1122 tp->t_srtt = TCPTV_SRTTBASE;
1123 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1124 tp->t_rttmin = tcp_rexmit_min;
1125 tp->t_rxtcur = TCPTV_RTOBASE;
1126 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1127 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1128 tp->t_rcvtime = ticks;
1129 /*
1130 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1131 * because the socket may be bound to an IPv6 wildcard address,
1132 * which may match an IPv4-mapped IPv6 address.
1133 */
1134 inp->inp_ip_ttl = V_ip_defttl;
1135 inp->inp_ppcb = tp;
1136#ifdef TCPPCAP
1137 /*
1138 * Init the TCP PCAP queues.
1139 */
1140 tcp_pcap_tcpcb_init(tp);
1141#endif
1142 return (tp); /* XXX */
1143}
1144
1145/*
1146 * Switch the congestion control algorithm back to NewReno for any active
1147 * control blocks using an algorithm which is about to go away.
1148 * This ensures the CC framework can allow the unload to proceed without leaving
1149 * any dangling pointers which would trigger a panic.
1150 * Returning non-zero would inform the CC framework that something went wrong
1151 * and it would be unsafe to allow the unload to proceed. However, there is no
1152 * way for this to occur with this implementation so we always return zero.
1153 */
1154int
1155tcp_ccalgounload(struct cc_algo *unload_algo)
1156{
1157 struct cc_algo *tmpalgo;
1158 struct inpcb *inp;
1159 struct tcpcb *tp;
1160 VNET_ITERATOR_DECL(vnet_iter);
1161
1162 /*
1163 * Check all active control blocks across all network stacks and change
1164 * any that are using "unload_algo" back to NewReno. If "unload_algo"
1165 * requires cleanup code to be run, call it.
1166 */
1167 VNET_LIST_RLOCK();
1168 VNET_FOREACH(vnet_iter) {
1169 CURVNET_SET(vnet_iter);
1170 INP_INFO_WLOCK(&V_tcbinfo);
1171 /*
1172 * New connections already part way through being initialised
1173 * with the CC algo we're removing will not race with this code
1174 * because the INP_INFO_WLOCK is held during initialisation. We
1175 * therefore don't enter the loop below until the connection
1176 * list has stabilised.
1177 */
1178 LIST_FOREACH(inp, &V_tcb, inp_list) {
1179 INP_WLOCK(inp);
1180 /* Important to skip tcptw structs. */
1181 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1182 (tp = intotcpcb(inp)) != NULL) {
1183 /*
1184 * By holding INP_WLOCK here, we are assured
1185 * that the connection is not currently
1186 * executing inside the CC module's functions
1187 * i.e. it is safe to make the switch back to
1188 * NewReno.
1189 */
1190 if (CC_ALGO(tp) == unload_algo) {
1191 tmpalgo = CC_ALGO(tp);
1192 /* NewReno does not require any init. */
1193 CC_ALGO(tp) = &newreno_cc_algo;
1194 if (tmpalgo->cb_destroy != NULL)
1195 tmpalgo->cb_destroy(tp->ccv);
1196 }
1197 }
1198 INP_WUNLOCK(inp);
1199 }
1200 INP_INFO_WUNLOCK(&V_tcbinfo);
1201 CURVNET_RESTORE();
1202 }
1203 VNET_LIST_RUNLOCK();
1204
1205 return (0);
1206}
1207
1208/*
1209 * Drop a TCP connection, reporting
1210 * the specified error. If connection is synchronized,
1211 * then send a RST to peer.
1212 */
1213struct tcpcb *
1214tcp_drop(struct tcpcb *tp, int errno)
1215{
1216 struct socket *so = tp->t_inpcb->inp_socket;
1217
1218 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1219 INP_WLOCK_ASSERT(tp->t_inpcb);
1220
1221 if (TCPS_HAVERCVDSYN(tp->t_state)) {
1222 tcp_state_change(tp, TCPS_CLOSED);
1223 (void) tp->t_fb->tfb_tcp_output(tp);
1224 TCPSTAT_INC(tcps_drops);
1225 } else
1226 TCPSTAT_INC(tcps_conndrops);
1227 if (errno == ETIMEDOUT && tp->t_softerror)
1228 errno = tp->t_softerror;
1229 so->so_error = errno;
1230 return (tcp_close(tp));
1231}
1232
1233void
1234tcp_discardcb(struct tcpcb *tp)
1235{
1236 struct inpcb *inp = tp->t_inpcb;
1237 struct socket *so = inp->inp_socket;
1238#ifdef INET6
1239 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1240#endif /* INET6 */
1241 int released;
1242
1243 INP_WLOCK_ASSERT(inp);
1244
1245 /*
1246 * Make sure that all of our timers are stopped before we delete the
1247 * PCB.
1248 *
1249 * If stopping a timer fails, we schedule a discard function in same
1250 * callout, and the last discard function called will take care of
1251 * deleting the tcpcb.
1252 */
1253 tcp_timer_stop(tp, TT_REXMT);
1254 tcp_timer_stop(tp, TT_PERSIST);
1255 tcp_timer_stop(tp, TT_KEEP);
1256 tcp_timer_stop(tp, TT_2MSL);
1257 tcp_timer_stop(tp, TT_DELACK);
1258 if (tp->t_fb->tfb_tcp_timer_stop_all) {
1259 /* Call the stop-all function of the methods */
1260 tp->t_fb->tfb_tcp_timer_stop_all(tp);
1261 }
1262
1263 /*
1264 * If we got enough samples through the srtt filter,
1265 * save the rtt and rttvar in the routing entry.
1266 * 'Enough' is arbitrarily defined as 4 rtt samples.
1267 * 4 samples is enough for the srtt filter to converge
1268 * to within enough % of the correct value; fewer samples
1269 * and we could save a bogus rtt. The danger is not high
1270 * as tcp quickly recovers from everything.
1271 * XXX: Works very well but needs some more statistics!
1272 */
1273 if (tp->t_rttupdated >= 4) {
1274 struct hc_metrics_lite metrics;
1275 u_long ssthresh;
1276
1277 bzero(&metrics, sizeof(metrics));
1278 /*
1279 * Update the ssthresh always when the conditions below
1280 * are satisfied. This gives us better new start value
1281 * for the congestion avoidance for new connections.
1282 * ssthresh is only set if packet loss occured on a session.
1283 *
1284 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1285 * being torn down. Ideally this code would not use 'so'.
1286 */
1287 ssthresh = tp->snd_ssthresh;
1288 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1289 /*
1290 * convert the limit from user data bytes to
1291 * packets then to packet data bytes.
1292 */
1293 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1294 if (ssthresh < 2)
1295 ssthresh = 2;
1296 ssthresh *= (u_long)(tp->t_maxseg +
1297#ifdef INET6
1298 (isipv6 ? sizeof (struct ip6_hdr) +
1299 sizeof (struct tcphdr) :
1300#endif
1301 sizeof (struct tcpiphdr)
1302#ifdef INET6
1303 )
1304#endif
1305 );
1306 } else
1307 ssthresh = 0;
1308 metrics.rmx_ssthresh = ssthresh;
1309
1310 metrics.rmx_rtt = tp->t_srtt;
1311 metrics.rmx_rttvar = tp->t_rttvar;
1312 metrics.rmx_cwnd = tp->snd_cwnd;
1313 metrics.rmx_sendpipe = 0;
1314 metrics.rmx_recvpipe = 0;
1315
1316 tcp_hc_update(&inp->inp_inc, &metrics);
1317 }
1318
1319 /* free the reassembly queue, if any */
1320 tcp_reass_flush(tp);
1321
1322#ifdef TCP_OFFLOAD
1323 /* Disconnect offload device, if any. */
1324 if (tp->t_flags & TF_TOE)
1325 tcp_offload_detach(tp);
1326#endif
1327
1328 tcp_free_sackholes(tp);
1329
1330#ifdef TCPPCAP
1331 /* Free the TCP PCAP queues. */
1332 tcp_pcap_drain(&(tp->t_inpkts));
1333 tcp_pcap_drain(&(tp->t_outpkts));
1334#endif
1335
1336 /* Allow the CC algorithm to clean up after itself. */
1337 if (CC_ALGO(tp)->cb_destroy != NULL)
1338 CC_ALGO(tp)->cb_destroy(tp->ccv);
1339
1340 khelp_destroy_osd(tp->osd);
1341
1342 CC_ALGO(tp) = NULL;
1343 inp->inp_ppcb = NULL;
1344 if ((tp->t_timers->tt_flags & TT_MASK) == 0) {
1345 /* We own the last reference on tcpcb, let's free it. */
1346 if ((tp->t_fb->tfb_tcp_timers_left) &&
1347 (tp->t_fb->tfb_tcp_timers_left(tp))) {
1348 /* Some fb timers left running! */
1349 return;
1350 }
1351 if (tp->t_fb->tfb_tcp_fb_fini)
1352 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1353 refcount_release(&tp->t_fb->tfb_refcnt);
1354 tp->t_inpcb = NULL;
1355 uma_zfree(V_tcpcb_zone, tp);
1356 released = in_pcbrele_wlocked(inp);
1357 KASSERT(!released, ("%s: inp %p should not have been released "
1358 "here", __func__, inp));
1359 }
1360}
1361
1362void
1363tcp_timer_2msl_discard(void *xtp)
1364{
1365
1366 tcp_timer_discard((struct tcpcb *)xtp, TT_2MSL);
1367}
1368
1369void
1370tcp_timer_keep_discard(void *xtp)
1371{
1372
1373 tcp_timer_discard((struct tcpcb *)xtp, TT_KEEP);
1374}
1375
1376void
1377tcp_timer_persist_discard(void *xtp)
1378{
1379
1380 tcp_timer_discard((struct tcpcb *)xtp, TT_PERSIST);
1381}
1382
1383void
1384tcp_timer_rexmt_discard(void *xtp)
1385{
1386
1387 tcp_timer_discard((struct tcpcb *)xtp, TT_REXMT);
1388}
1389
1390void
1391tcp_timer_delack_discard(void *xtp)
1392{
1393
1394 tcp_timer_discard((struct tcpcb *)xtp, TT_DELACK);
1395}
1396
1397void
1398tcp_timer_discard(struct tcpcb *tp, uint32_t timer_type)
1399{
1400 struct inpcb *inp;
1401
1402 CURVNET_SET(tp->t_vnet);
1403 INP_INFO_RLOCK(&V_tcbinfo);
1404 inp = tp->t_inpcb;
1405 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1406 __func__, tp));
1407 INP_WLOCK(inp);
1408 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1409 ("%s: tcpcb has to be stopped here", __func__));
1410 KASSERT((tp->t_timers->tt_flags & timer_type) != 0,
1411 ("%s: discard callout should be running", __func__));
1412 tp->t_timers->tt_flags &= ~timer_type;
1413 if ((tp->t_timers->tt_flags & TT_MASK) == 0) {
1414 /* We own the last reference on this tcpcb, let's free it. */
1415 if ((tp->t_fb->tfb_tcp_timers_left) &&
1416 (tp->t_fb->tfb_tcp_timers_left(tp))) {
1417 /* Some fb timers left running! */
1418 goto leave;
1419 }
1420 if (tp->t_fb->tfb_tcp_fb_fini)
1421 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1422 refcount_release(&tp->t_fb->tfb_refcnt);
1423 tp->t_inpcb = NULL;
1424 uma_zfree(V_tcpcb_zone, tp);
1425 if (in_pcbrele_wlocked(inp)) {
1426 INP_INFO_RUNLOCK(&V_tcbinfo);
1427 CURVNET_RESTORE();
1428 return;
1429 }
1430 }
1431leave:
1432 INP_WUNLOCK(inp);
1433 INP_INFO_RUNLOCK(&V_tcbinfo);
1434 CURVNET_RESTORE();
1435}
1436
1437/*
1438 * Attempt to close a TCP control block, marking it as dropped, and freeing
1439 * the socket if we hold the only reference.
1440 */
1441struct tcpcb *
1442tcp_close(struct tcpcb *tp)
1443{
1444 struct inpcb *inp = tp->t_inpcb;
1445 struct socket *so;
1446
1447 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1448 INP_WLOCK_ASSERT(inp);
1449
1450#ifdef TCP_OFFLOAD
1451 if (tp->t_state == TCPS_LISTEN)
1452 tcp_offload_listen_stop(tp);
1453#endif
1454#ifdef TCP_RFC7413
1455 /*
1456 * This releases the TFO pending counter resource for TFO listen
1457 * sockets as well as passively-created TFO sockets that transition
1458 * from SYN_RECEIVED to CLOSED.
1459 */
1460 if (tp->t_tfo_pending) {
1461 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
1462 tp->t_tfo_pending = NULL;
1463 }
1464#endif
1465 in_pcbdrop(inp);
1466 TCPSTAT_INC(tcps_closed);
1467 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1468 so = inp->inp_socket;
1469 soisdisconnected(so);
1470 if (inp->inp_flags & INP_SOCKREF) {
1471 KASSERT(so->so_state & SS_PROTOREF,
1472 ("tcp_close: !SS_PROTOREF"));
1473 inp->inp_flags &= ~INP_SOCKREF;
1474 INP_WUNLOCK(inp);
1475 ACCEPT_LOCK();
1476 SOCK_LOCK(so);
1477 so->so_state &= ~SS_PROTOREF;
1478 sofree(so);
1479 return (NULL);
1480 }
1481 return (tp);
1482}
1483
1484void
1485tcp_drain(void)
1486{
1487 VNET_ITERATOR_DECL(vnet_iter);
1488
1489 if (!do_tcpdrain)
1490 return;
1491
1492 VNET_LIST_RLOCK_NOSLEEP();
1493 VNET_FOREACH(vnet_iter) {
1494 CURVNET_SET(vnet_iter);
1495 struct inpcb *inpb;
1496 struct tcpcb *tcpb;
1497
1498 /*
1499 * Walk the tcpbs, if existing, and flush the reassembly queue,
1500 * if there is one...
1501 * XXX: The "Net/3" implementation doesn't imply that the TCP
1502 * reassembly queue should be flushed, but in a situation
1503 * where we're really low on mbufs, this is potentially
1504 * useful.
1505 */
1506 INP_INFO_WLOCK(&V_tcbinfo);
1507 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1508 if (inpb->inp_flags & INP_TIMEWAIT)
1509 continue;
1510 INP_WLOCK(inpb);
1511 if ((tcpb = intotcpcb(inpb)) != NULL) {
1512 tcp_reass_flush(tcpb);
1513 tcp_clean_sackreport(tcpb);
1514 }
1515 INP_WUNLOCK(inpb);
1516 }
1517 INP_INFO_WUNLOCK(&V_tcbinfo);
1518 CURVNET_RESTORE();
1519 }
1520 VNET_LIST_RUNLOCK_NOSLEEP();
1521}
1522
1523/*
1524 * Notify a tcp user of an asynchronous error;
1525 * store error as soft error, but wake up user
1526 * (for now, won't do anything until can select for soft error).
1527 *
1528 * Do not wake up user since there currently is no mechanism for
1529 * reporting soft errors (yet - a kqueue filter may be added).
1530 */
1531static struct inpcb *
1532tcp_notify(struct inpcb *inp, int error)
1533{
1534 struct tcpcb *tp;
1535
1536 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1537 INP_WLOCK_ASSERT(inp);
1538
1539 if ((inp->inp_flags & INP_TIMEWAIT) ||
1540 (inp->inp_flags & INP_DROPPED))
1541 return (inp);
1542
1543 tp = intotcpcb(inp);
1544 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1545
1546 /*
1547 * Ignore some errors if we are hooked up.
1548 * If connection hasn't completed, has retransmitted several times,
1549 * and receives a second error, give up now. This is better
1550 * than waiting a long time to establish a connection that
1551 * can never complete.
1552 */
1553 if (tp->t_state == TCPS_ESTABLISHED &&
1554 (error == EHOSTUNREACH || error == ENETUNREACH ||
1555 error == EHOSTDOWN)) {
1556 return (inp);
1557 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1558 tp->t_softerror) {
1559 tp = tcp_drop(tp, error);
1560 if (tp != NULL)
1561 return (inp);
1562 else
1563 return (NULL);
1564 } else {
1565 tp->t_softerror = error;
1566 return (inp);
1567 }
1568#if 0
1569 wakeup( &so->so_timeo);
1570 sorwakeup(so);
1571 sowwakeup(so);
1572#endif
1573}
1574
1575static int
1576tcp_pcblist(SYSCTL_HANDLER_ARGS)
1577{
1578 int error, i, m, n, pcb_count;
1579 struct inpcb *inp, **inp_list;
1580 inp_gen_t gencnt;
1581 struct xinpgen xig;
1582
1583 /*
1584 * The process of preparing the TCB list is too time-consuming and
1585 * resource-intensive to repeat twice on every request.
1586 */
1587 if (req->oldptr == NULL) {
1588 n = V_tcbinfo.ipi_count + syncache_pcbcount();
1589 n += imax(n / 8, 10);
1590 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1591 return (0);
1592 }
1593
1594 if (req->newptr != NULL)
1595 return (EPERM);
1596
1597 /*
1598 * OK, now we're committed to doing something.
1599 */
1600 INP_LIST_RLOCK(&V_tcbinfo);
1601 gencnt = V_tcbinfo.ipi_gencnt;
1602 n = V_tcbinfo.ipi_count;
1603 INP_LIST_RUNLOCK(&V_tcbinfo);
1604
1605 m = syncache_pcbcount();
1606
1607 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1608 + (n + m) * sizeof(struct xtcpcb));
1609 if (error != 0)
1610 return (error);
1611
1612 xig.xig_len = sizeof xig;
1613 xig.xig_count = n + m;
1614 xig.xig_gen = gencnt;
1615 xig.xig_sogen = so_gencnt;
1616 error = SYSCTL_OUT(req, &xig, sizeof xig);
1617 if (error)
1618 return (error);
1619
1620 error = syncache_pcblist(req, m, &pcb_count);
1621 if (error)
1622 return (error);
1623
1624 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1625 if (inp_list == NULL)
1626 return (ENOMEM);
1627
1628 INP_INFO_WLOCK(&V_tcbinfo);
1629 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1630 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1631 INP_WLOCK(inp);
1632 if (inp->inp_gencnt <= gencnt) {
1633 /*
1634 * XXX: This use of cr_cansee(), introduced with
1635 * TCP state changes, is not quite right, but for
1636 * now, better than nothing.
1637 */
1638 if (inp->inp_flags & INP_TIMEWAIT) {
1639 if (intotw(inp) != NULL)
1640 error = cr_cansee(req->td->td_ucred,
1641 intotw(inp)->tw_cred);
1642 else
1643 error = EINVAL; /* Skip this inp. */
1644 } else
1645 error = cr_canseeinpcb(req->td->td_ucred, inp);
1646 if (error == 0) {
1647 in_pcbref(inp);
1648 inp_list[i++] = inp;
1649 }
1650 }
1651 INP_WUNLOCK(inp);
1652 }
1653 INP_INFO_WUNLOCK(&V_tcbinfo);
1654 n = i;
1655
1656 error = 0;
1657 for (i = 0; i < n; i++) {
1658 inp = inp_list[i];
1659 INP_RLOCK(inp);
1660 if (inp->inp_gencnt <= gencnt) {
1661 struct xtcpcb xt;
1662 void *inp_ppcb;
1663
1664 bzero(&xt, sizeof(xt));
1665 xt.xt_len = sizeof xt;
1666 /* XXX should avoid extra copy */
1667 bcopy(inp, &xt.xt_inp, sizeof *inp);
1668 inp_ppcb = inp->inp_ppcb;
1669 if (inp_ppcb == NULL)
1670 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1671 else if (inp->inp_flags & INP_TIMEWAIT) {
1672 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1673 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1674 } else {
1675 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1676 if (xt.xt_tp.t_timers)
1677 tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1678 }
1679 if (inp->inp_socket != NULL)
1680 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1681 else {
1682 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1683 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1684 }
1685 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1686 INP_RUNLOCK(inp);
1687 error = SYSCTL_OUT(req, &xt, sizeof xt);
1688 } else
1689 INP_RUNLOCK(inp);
1690 }
1691 INP_INFO_RLOCK(&V_tcbinfo);
1692 for (i = 0; i < n; i++) {
1693 inp = inp_list[i];
1694 INP_RLOCK(inp);
1695 if (!in_pcbrele_rlocked(inp))
1696 INP_RUNLOCK(inp);
1697 }
1698 INP_INFO_RUNLOCK(&V_tcbinfo);
1699
1700 if (!error) {
1701 /*
1702 * Give the user an updated idea of our state.
1703 * If the generation differs from what we told
1704 * her before, she knows that something happened
1705 * while we were processing this request, and it
1706 * might be necessary to retry.
1707 */
1708 INP_LIST_RLOCK(&V_tcbinfo);
1709 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1710 xig.xig_sogen = so_gencnt;
1711 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1712 INP_LIST_RUNLOCK(&V_tcbinfo);
1713 error = SYSCTL_OUT(req, &xig, sizeof xig);
1714 }
1715 free(inp_list, M_TEMP);
1716 return (error);
1717}
1718
1719SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1720 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1721 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1722
1723#ifdef INET
1724static int
1725tcp_getcred(SYSCTL_HANDLER_ARGS)
1726{
1727 struct xucred xuc;
1728 struct sockaddr_in addrs[2];
1729 struct inpcb *inp;
1730 int error;
1731
1732 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1733 if (error)
1734 return (error);
1735 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1736 if (error)
1737 return (error);
1738 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1739 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1740 if (inp != NULL) {
1741 if (inp->inp_socket == NULL)
1742 error = ENOENT;
1743 if (error == 0)
1744 error = cr_canseeinpcb(req->td->td_ucred, inp);
1745 if (error == 0)
1746 cru2x(inp->inp_cred, &xuc);
1747 INP_RUNLOCK(inp);
1748 } else
1749 error = ENOENT;
1750 if (error == 0)
1751 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1752 return (error);
1753}
1754
1755SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1756 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1757 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1758#endif /* INET */
1759
1760#ifdef INET6
1761static int
1762tcp6_getcred(SYSCTL_HANDLER_ARGS)
1763{
1764 struct xucred xuc;
1765 struct sockaddr_in6 addrs[2];
1766 struct inpcb *inp;
1767 int error;
1768#ifdef INET
1769 int mapped = 0;
1770#endif
1771
1772 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1773 if (error)
1774 return (error);
1775 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1776 if (error)
1777 return (error);
1778 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1779 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1780 return (error);
1781 }
1782 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1783#ifdef INET
1784 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1785 mapped = 1;
1786 else
1787#endif
1788 return (EINVAL);
1789 }
1790
1791#ifdef INET
1792 if (mapped == 1)
1793 inp = in_pcblookup(&V_tcbinfo,
1794 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1795 addrs[1].sin6_port,
1796 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1797 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1798 else
1799#endif
1800 inp = in6_pcblookup(&V_tcbinfo,
1801 &addrs[1].sin6_addr, addrs[1].sin6_port,
1802 &addrs[0].sin6_addr, addrs[0].sin6_port,
1803 INPLOOKUP_RLOCKPCB, NULL);
1804 if (inp != NULL) {
1805 if (inp->inp_socket == NULL)
1806 error = ENOENT;
1807 if (error == 0)
1808 error = cr_canseeinpcb(req->td->td_ucred, inp);
1809 if (error == 0)
1810 cru2x(inp->inp_cred, &xuc);
1811 INP_RUNLOCK(inp);
1812 } else
1813 error = ENOENT;
1814 if (error == 0)
1815 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1816 return (error);
1817}
1818
1819SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1820 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1821 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1822#endif /* INET6 */
1823
1824
1825#ifdef INET
1826void
1827tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1828{
1829 struct ip *ip = vip;
1830 struct tcphdr *th;
1831 struct in_addr faddr;
1832 struct inpcb *inp;
1833 struct tcpcb *tp;
1834 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1835 struct icmp *icp;
1836 struct in_conninfo inc;
1837 tcp_seq icmp_tcp_seq;
1838 int mtu;
1839
1840 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1841 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1842 return;
1843
1844 if (cmd == PRC_MSGSIZE)
1845 notify = tcp_mtudisc_notify;
1846 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1847 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1848 notify = tcp_drop_syn_sent;
1849 /*
1850 * Redirects don't need to be handled up here.
1851 */
1852 else if (PRC_IS_REDIRECT(cmd))
1853 return;
1854 /*
1855 * Hostdead is ugly because it goes linearly through all PCBs.
1856 * XXX: We never get this from ICMP, otherwise it makes an
1857 * excellent DoS attack on machines with many connections.
1858 */
1859 else if (cmd == PRC_HOSTDEAD)
1860 ip = NULL;
1861 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1862 return;
1863
1864 if (ip == NULL) {
1865 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1866 return;
1867 }
1868
1869 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
1870 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1871 INP_INFO_RLOCK(&V_tcbinfo);
1872 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
1873 th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1874 if (inp != NULL) {
1875 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1876 !(inp->inp_flags & INP_DROPPED) &&
1877 !(inp->inp_socket == NULL)) {
1878 icmp_tcp_seq = ntohl(th->th_seq);
1879 tp = intotcpcb(inp);
1880 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1881 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1882 if (cmd == PRC_MSGSIZE) {
1883 /*
1884 * MTU discovery:
1885 * If we got a needfrag set the MTU
1886 * in the route to the suggested new
1887 * value (if given) and then notify.
1888 */
1889 mtu = ntohs(icp->icmp_nextmtu);
1890 /*
1891 * If no alternative MTU was
1892 * proposed, try the next smaller
1893 * one.
1894 */
1895 if (!mtu)
1896 mtu = ip_next_mtu(
1897 ntohs(ip->ip_len), 1);
1898 if (mtu < V_tcp_minmss +
1899 sizeof(struct tcpiphdr))
1900 mtu = V_tcp_minmss +
1901 sizeof(struct tcpiphdr);
1902 /*
1903 * Only process the offered MTU if it
1904 * is smaller than the current one.
1905 */
1906 if (mtu < tp->t_maxseg +
1907 sizeof(struct tcpiphdr)) {
1908 bzero(&inc, sizeof(inc));
1909 inc.inc_faddr = faddr;
1910 inc.inc_fibnum =
1911 inp->inp_inc.inc_fibnum;
1912 tcp_hc_updatemtu(&inc, mtu);
1913 tcp_mtudisc(inp, mtu);
1914 }
1915 } else
1916 inp = (*notify)(inp,
1917 inetctlerrmap[cmd]);
1918 }
1919 }
1920 if (inp != NULL)
1921 INP_WUNLOCK(inp);
1922 } else {
1923 bzero(&inc, sizeof(inc));
1924 inc.inc_fport = th->th_dport;
1925 inc.inc_lport = th->th_sport;
1926 inc.inc_faddr = faddr;
1927 inc.inc_laddr = ip->ip_src;
1928 syncache_unreach(&inc, th);
1929 }
1930 INP_INFO_RUNLOCK(&V_tcbinfo);
1931}
1932#endif /* INET */
1933
1934#ifdef INET6
1935void
1936tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1937{
1938 struct tcphdr th;
1939 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1940 struct ip6_hdr *ip6;
1941 struct mbuf *m;
1942 struct ip6ctlparam *ip6cp = NULL;
1943 const struct sockaddr_in6 *sa6_src = NULL;
1944 int off;
1945 struct tcp_portonly {
1946 u_int16_t th_sport;
1947 u_int16_t th_dport;
1948 } *thp;
1949
1950 if (sa->sa_family != AF_INET6 ||
1951 sa->sa_len != sizeof(struct sockaddr_in6))
1952 return;
1953
1954 if (cmd == PRC_MSGSIZE)
1955 notify = tcp_mtudisc_notify;
1956 else if (!PRC_IS_REDIRECT(cmd) &&
1957 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1958 return;
1959
1960 /* if the parameter is from icmp6, decode it. */
1961 if (d != NULL) {
1962 ip6cp = (struct ip6ctlparam *)d;
1963 m = ip6cp->ip6c_m;
1964 ip6 = ip6cp->ip6c_ip6;
1965 off = ip6cp->ip6c_off;
1966 sa6_src = ip6cp->ip6c_src;
1967 } else {
1968 m = NULL;
1969 ip6 = NULL;
1970 off = 0; /* fool gcc */
1971 sa6_src = &sa6_any;
1972 }
1973
1974 if (ip6 != NULL) {
1975 struct in_conninfo inc;
1976 /*
1977 * XXX: We assume that when IPV6 is non NULL,
1978 * M and OFF are valid.
1979 */
1980
1981 /* check if we can safely examine src and dst ports */
1982 if (m->m_pkthdr.len < off + sizeof(*thp))
1983 return;
1984
1985 bzero(&th, sizeof(th));
1986 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1987
1988 in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1989 (struct sockaddr *)ip6cp->ip6c_src,
1990 th.th_sport, cmd, NULL, notify);
1991
1992 bzero(&inc, sizeof(inc));
1993 inc.inc_fport = th.th_dport;
1994 inc.inc_lport = th.th_sport;
1995 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1996 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1997 inc.inc_flags |= INC_ISIPV6;
1998 INP_INFO_RLOCK(&V_tcbinfo);
1999 syncache_unreach(&inc, &th);
2000 INP_INFO_RUNLOCK(&V_tcbinfo);
2001 } else
2002 in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
2003 0, cmd, NULL, notify);
2004}
2005#endif /* INET6 */
2006
2007
2008/*
2009 * Following is where TCP initial sequence number generation occurs.
2010 *
2011 * There are two places where we must use initial sequence numbers:
2012 * 1. In SYN-ACK packets.
2013 * 2. In SYN packets.
2014 *
2015 * All ISNs for SYN-ACK packets are generated by the syncache. See
2016 * tcp_syncache.c for details.
2017 *
2018 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2019 * depends on this property. In addition, these ISNs should be
2020 * unguessable so as to prevent connection hijacking. To satisfy
2021 * the requirements of this situation, the algorithm outlined in
2022 * RFC 1948 is used, with only small modifications.
2023 *
2024 * Implementation details:
2025 *
2026 * Time is based off the system timer, and is corrected so that it
2027 * increases by one megabyte per second. This allows for proper
2028 * recycling on high speed LANs while still leaving over an hour
2029 * before rollover.
2030 *
2031 * As reading the *exact* system time is too expensive to be done
2032 * whenever setting up a TCP connection, we increment the time
2033 * offset in two ways. First, a small random positive increment
2034 * is added to isn_offset for each connection that is set up.
2035 * Second, the function tcp_isn_tick fires once per clock tick
2036 * and increments isn_offset as necessary so that sequence numbers
2037 * are incremented at approximately ISN_BYTES_PER_SECOND. The
2038 * random positive increments serve only to ensure that the same
2039 * exact sequence number is never sent out twice (as could otherwise
2040 * happen when a port is recycled in less than the system tick
2041 * interval.)
2042 *
2043 * net.inet.tcp.isn_reseed_interval controls the number of seconds
2044 * between seeding of isn_secret. This is normally set to zero,
2045 * as reseeding should not be necessary.
2046 *
2047 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2048 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
2049 * general, this means holding an exclusive (write) lock.
2050 */
2051
2052#define ISN_BYTES_PER_SECOND 1048576
2053#define ISN_STATIC_INCREMENT 4096
2054#define ISN_RANDOM_INCREMENT (4096 - 1)
2055
2056static VNET_DEFINE(u_char, isn_secret[32]);
2057static VNET_DEFINE(int, isn_last);
2058static VNET_DEFINE(int, isn_last_reseed);
2059static VNET_DEFINE(u_int32_t, isn_offset);
2060static VNET_DEFINE(u_int32_t, isn_offset_old);
2061
2062#define V_isn_secret VNET(isn_secret)
2063#define V_isn_last VNET(isn_last)
2064#define V_isn_last_reseed VNET(isn_last_reseed)
2065#define V_isn_offset VNET(isn_offset)
2066#define V_isn_offset_old VNET(isn_offset_old)
2067
2068tcp_seq
2069tcp_new_isn(struct tcpcb *tp)
2070{
2071 MD5_CTX isn_ctx;
2072 u_int32_t md5_buffer[4];
2073 tcp_seq new_isn;
2074 u_int32_t projected_offset;
2075
2076 INP_WLOCK_ASSERT(tp->t_inpcb);
2077
2078 ISN_LOCK();
2079 /* Seed if this is the first use, reseed if requested. */
2080 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2081 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2082 < (u_int)ticks))) {
2083 read_random(&V_isn_secret, sizeof(V_isn_secret));
2084 V_isn_last_reseed = ticks;
2085 }
2086
2087 /* Compute the md5 hash and return the ISN. */
2088 MD5Init(&isn_ctx);
2089 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
2090 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
2091#ifdef INET6
2092 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
2093 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
2094 sizeof(struct in6_addr));
2095 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
2096 sizeof(struct in6_addr));
2097 } else
2098#endif
2099 {
2100 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
2101 sizeof(struct in_addr));
2102 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
2103 sizeof(struct in_addr));
2104 }
2105 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
2106 MD5Final((u_char *) &md5_buffer, &isn_ctx);
2107 new_isn = (tcp_seq) md5_buffer[0];
2108 V_isn_offset += ISN_STATIC_INCREMENT +
2109 (arc4random() & ISN_RANDOM_INCREMENT);
2110 if (ticks != V_isn_last) {
2111 projected_offset = V_isn_offset_old +
2112 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2113 if (SEQ_GT(projected_offset, V_isn_offset))
2114 V_isn_offset = projected_offset;
2115 V_isn_offset_old = V_isn_offset;
2116 V_isn_last = ticks;
2117 }
2118 new_isn += V_isn_offset;
2119 ISN_UNLOCK();
2120 return (new_isn);
2121}
2122
2123/*
2124 * When a specific ICMP unreachable message is received and the
2125 * connection state is SYN-SENT, drop the connection. This behavior
2126 * is controlled by the icmp_may_rst sysctl.
2127 */
2128struct inpcb *
2129tcp_drop_syn_sent(struct inpcb *inp, int errno)
2130{
2131 struct tcpcb *tp;
2132
2133 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2134 INP_WLOCK_ASSERT(inp);
2135
2136 if ((inp->inp_flags & INP_TIMEWAIT) ||
2137 (inp->inp_flags & INP_DROPPED))
2138 return (inp);
2139
2140 tp = intotcpcb(inp);
2141 if (tp->t_state != TCPS_SYN_SENT)
2142 return (inp);
2143
2144 tp = tcp_drop(tp, errno);
2145 if (tp != NULL)
2146 return (inp);
2147 else
2148 return (NULL);
2149}
2150
2151/*
2152 * When `need fragmentation' ICMP is received, update our idea of the MSS
2153 * based on the new value. Also nudge TCP to send something, since we
2154 * know the packet we just sent was dropped.
2155 * This duplicates some code in the tcp_mss() function in tcp_input.c.
2156 */
2157static struct inpcb *
2158tcp_mtudisc_notify(struct inpcb *inp, int error)
2159{
2160
2161 tcp_mtudisc(inp, -1);
2162 return (inp);
2163}
2164
2165static void
2166tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2167{
2168 struct tcpcb *tp;
2169 struct socket *so;
2170
2171 INP_WLOCK_ASSERT(inp);
2172 if ((inp->inp_flags & INP_TIMEWAIT) ||
2173 (inp->inp_flags & INP_DROPPED))
2174 return;
2175
2176 tp = intotcpcb(inp);
2177 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2178
2179 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2180
2181 so = inp->inp_socket;
2182 SOCKBUF_LOCK(&so->so_snd);
2183 /* If the mss is larger than the socket buffer, decrease the mss. */
2184 if (so->so_snd.sb_hiwat < tp->t_maxseg)
2185 tp->t_maxseg = so->so_snd.sb_hiwat;
2186 SOCKBUF_UNLOCK(&so->so_snd);
2187
2188 TCPSTAT_INC(tcps_mturesent);
2189 tp->t_rtttime = 0;
2190 tp->snd_nxt = tp->snd_una;
2191 tcp_free_sackholes(tp);
2192 tp->snd_recover = tp->snd_max;
2193 if (tp->t_flags & TF_SACK_PERMIT)
2194 EXIT_FASTRECOVERY(tp->t_flags);
2195 tp->t_fb->tfb_tcp_output(tp);
2196}
2197
2198#ifdef INET
2199/*
2200 * Look-up the routing entry to the peer of this inpcb. If no route
2201 * is found and it cannot be allocated, then return 0. This routine
2202 * is called by TCP routines that access the rmx structure and by
2203 * tcp_mss_update to get the peer/interface MTU.
2204 */
2205u_long
2206tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2207{
2208 struct route sro;
2209 struct sockaddr_in *dst;
2210 struct ifnet *ifp;
2211 u_long maxmtu = 0;
2212
2213 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2214
2215 bzero(&sro, sizeof(sro));
2216 if (inc->inc_faddr.s_addr != INADDR_ANY) {
2217 dst = (struct sockaddr_in *)&sro.ro_dst;
2218 dst->sin_family = AF_INET;
2219 dst->sin_len = sizeof(*dst);
2220 dst->sin_addr = inc->inc_faddr;
2221 in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
2222 }
2223 if (sro.ro_rt != NULL) {
2224 ifp = sro.ro_rt->rt_ifp;
2225 if (sro.ro_rt->rt_mtu == 0)
2226 maxmtu = ifp->if_mtu;
2227 else
2228 maxmtu = min(sro.ro_rt->rt_mtu, ifp->if_mtu);
2229
2230 /* Report additional interface capabilities. */
2231 if (cap != NULL) {
2232 if (ifp->if_capenable & IFCAP_TSO4 &&
2233 ifp->if_hwassist & CSUM_TSO) {
2234 cap->ifcap |= CSUM_TSO;
2235 cap->tsomax = ifp->if_hw_tsomax;
2236 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2237 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2238 }
2239 }
2240 RTFREE(sro.ro_rt);
2241 }
2242 return (maxmtu);
2243}
2244#endif /* INET */
2245
2246#ifdef INET6
2247u_long
2248tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2249{
2250 struct route_in6 sro6;
2251 struct ifnet *ifp;
2252 u_long maxmtu = 0;
2253
2254 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2255
2256 bzero(&sro6, sizeof(sro6));
2257 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2258 sro6.ro_dst.sin6_family = AF_INET6;
2259 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
2260 sro6.ro_dst.sin6_addr = inc->inc6_faddr;
2261 in6_rtalloc_ign(&sro6, 0, inc->inc_fibnum);
2262 }
2263 if (sro6.ro_rt != NULL) {
2264 ifp = sro6.ro_rt->rt_ifp;
2265 if (sro6.ro_rt->rt_mtu == 0)
2266 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
2267 else
2268 maxmtu = min(sro6.ro_rt->rt_mtu,
2269 IN6_LINKMTU(sro6.ro_rt->rt_ifp));
2270
2271 /* Report additional interface capabilities. */
2272 if (cap != NULL) {
2273 if (ifp->if_capenable & IFCAP_TSO6 &&
2274 ifp->if_hwassist & CSUM_TSO) {
2275 cap->ifcap |= CSUM_TSO;
2276 cap->tsomax = ifp->if_hw_tsomax;
2277 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2278 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2279 }
2280 }
2281 RTFREE(sro6.ro_rt);
2282 }
2283
2284 return (maxmtu);
2285}
2286#endif /* INET6 */
2287
2288/*
2289 * Calculate effective SMSS per RFC5681 definition for a given TCP
2290 * connection at its current state, taking into account SACK and etc.
2291 */
2292u_int
2293tcp_maxseg(const struct tcpcb *tp)
2294{
2295 u_int optlen;
2296
2297 if (tp->t_flags & TF_NOOPT)
2298 return (tp->t_maxseg);
2299
2300 /*
2301 * Here we have a simplified code from tcp_addoptions(),
2302 * without a proper loop, and having most of paddings hardcoded.
2303 * We might make mistakes with padding here in some edge cases,
2304 * but this is harmless, since result of tcp_maxseg() is used
2305 * only in cwnd and ssthresh estimations.
2306 */
2307#define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4)
2308 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2309 if (tp->t_flags & TF_RCVD_TSTMP)
2310 optlen = TCPOLEN_TSTAMP_APPA;
2311 else
2312 optlen = 0;
2313#ifdef TCP_SIGNATURE
2314 if (tp->t_flags & TF_SIGNATURE)
2315 optlen += PAD(TCPOLEN_SIGNATURE);
2316#endif
2317 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
2318 optlen += TCPOLEN_SACKHDR;
2319 optlen += tp->rcv_numsacks * TCPOLEN_SACK;
2320 optlen = PAD(optlen);
2321 }
2322 } else {
2323 if (tp->t_flags & TF_REQ_TSTMP)
2324 optlen = TCPOLEN_TSTAMP_APPA;
2325 else
2326 optlen = PAD(TCPOLEN_MAXSEG);
2327 if (tp->t_flags & TF_REQ_SCALE)
2328 optlen += PAD(TCPOLEN_WINDOW);
2329#ifdef TCP_SIGNATURE
2330 if (tp->t_flags & TF_SIGNATURE)
2331 optlen += PAD(TCPOLEN_SIGNATURE);
2332#endif
2333 if (tp->t_flags & TF_SACK_PERMIT)
2334 optlen += PAD(TCPOLEN_SACK_PERMITTED);
2335 }
2336#undef PAD
2337 optlen = min(optlen, TCP_MAXOLEN);
2338 return (tp->t_maxseg - optlen);
2339}
2340
2341#ifdef IPSEC
2342/* compute ESP/AH header size for TCP, including outer IP header. */
2343size_t
2344ipsec_hdrsiz_tcp(struct tcpcb *tp)
2345{
2346 struct inpcb *inp;
2347 struct mbuf *m;
2348 size_t hdrsiz;
2349 struct ip *ip;
2350#ifdef INET6
2351 struct ip6_hdr *ip6;
2352#endif
2353 struct tcphdr *th;
2354
2355 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL) ||
2356 (!key_havesp(IPSEC_DIR_OUTBOUND)))
2357 return (0);
2358 m = m_gethdr(M_NOWAIT, MT_DATA);
2359 if (!m)
2360 return (0);
2361
2362#ifdef INET6
2363 if ((inp->inp_vflag & INP_IPV6) != 0) {
2364 ip6 = mtod(m, struct ip6_hdr *);
2365 th = (struct tcphdr *)(ip6 + 1);
2366 m->m_pkthdr.len = m->m_len =
2367 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
2368 tcpip_fillheaders(inp, ip6, th);
2369 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2370 } else
2371#endif /* INET6 */
2372 {
2373 ip = mtod(m, struct ip *);
2374 th = (struct tcphdr *)(ip + 1);
2375 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
2376 tcpip_fillheaders(inp, ip, th);
2377 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2378 }
2379
2380 m_free(m);
2381 return (hdrsiz);
2382}
2383#endif /* IPSEC */
2384
2385#ifdef TCP_SIGNATURE
2386/*
2387 * Callback function invoked by m_apply() to digest TCP segment data
2388 * contained within an mbuf chain.
2389 */
2390static int
2391tcp_signature_apply(void *fstate, void *data, u_int len)
2392{
2393
2394 MD5Update(fstate, (u_char *)data, len);
2395 return (0);
2396}
2397
2398/*
2399 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
2400 * search with the destination IP address, and a 'magic SPI' to be
2401 * determined by the application. This is hardcoded elsewhere to 1179
2402*/
2403struct secasvar *
2404tcp_get_sav(struct mbuf *m, u_int direction)
2405{
2406 union sockaddr_union dst;
2407 struct secasvar *sav;
2408 struct ip *ip;
2409#ifdef INET6
2410 struct ip6_hdr *ip6;
2411 char ip6buf[INET6_ADDRSTRLEN];
2412#endif
2413
2414 /* Extract the destination from the IP header in the mbuf. */
2415 bzero(&dst, sizeof(union sockaddr_union));
2416 ip = mtod(m, struct ip *);
2417#ifdef INET6
2418 ip6 = NULL; /* Make the compiler happy. */
2419#endif
2420 switch (ip->ip_v) {
2421#ifdef INET
2422 case IPVERSION:
2423 dst.sa.sa_len = sizeof(struct sockaddr_in);
2424 dst.sa.sa_family = AF_INET;
2425 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
2426 ip->ip_src : ip->ip_dst;
2427 break;
2428#endif
2429#ifdef INET6
2430 case (IPV6_VERSION >> 4):
2431 ip6 = mtod(m, struct ip6_hdr *);
2432 dst.sa.sa_len = sizeof(struct sockaddr_in6);
2433 dst.sa.sa_family = AF_INET6;
2434 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
2435 ip6->ip6_src : ip6->ip6_dst;
2436 break;
2437#endif
2438 default:
2439 return (NULL);
2440 /* NOTREACHED */
2441 break;
2442 }
2443
2444 /* Look up an SADB entry which matches the address of the peer. */
2445 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2446 if (sav == NULL) {
2447 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
2448 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
2449#ifdef INET6
2450 (ip->ip_v == (IPV6_VERSION >> 4)) ?
2451 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2452#endif
2453 "(unsupported)"));
2454 }
2455
2456 return (sav);
2457}
2458
2459/*
2460 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2461 *
2462 * Parameters:
2463 * m pointer to head of mbuf chain
2464 * len length of TCP segment data, excluding options
2465 * optlen length of TCP segment options
2466 * buf pointer to storage for computed MD5 digest
2467 * sav pointer to security assosiation
2468 *
2469 * We do this over ip, tcphdr, segment data, and the key in the SADB.
2470 * When called from tcp_input(), we can be sure that th_sum has been
2471 * zeroed out and verified already.
2472 *
2473 * Releases reference to SADB key before return.
2474 *
2475 * Return 0 if successful, otherwise return -1.
2476 *
2477 */
2478int
2479tcp_signature_do_compute(struct mbuf *m, int len, int optlen,
2480 u_char *buf, struct secasvar *sav)
2481{
2482#ifdef INET
2483 struct ippseudo ippseudo;
2484#endif
2485 MD5_CTX ctx;
2486 int doff;
2487 struct ip *ip;
2488#ifdef INET
2489 struct ipovly *ipovly;
2490#endif
2491 struct tcphdr *th;
2492#ifdef INET6
2493 struct ip6_hdr *ip6;
2494 struct in6_addr in6;
2495 uint32_t plen;
2496 uint16_t nhdr;
2497#endif
2498 u_short savecsum;
2499
2500 KASSERT(m != NULL, ("NULL mbuf chain"));
2501 KASSERT(buf != NULL, ("NULL signature pointer"));
2502
2503 /* Extract the destination from the IP header in the mbuf. */
2504 ip = mtod(m, struct ip *);
2505#ifdef INET6
2506 ip6 = NULL; /* Make the compiler happy. */
2507#endif
2508
2509 MD5Init(&ctx);
2510 /*
2511 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2512 *
2513 * XXX The ippseudo header MUST be digested in network byte order,
2514 * or else we'll fail the regression test. Assume all fields we've
2515 * been doing arithmetic on have been in host byte order.
2516 * XXX One cannot depend on ipovly->ih_len here. When called from
2517 * tcp_output(), the underlying ip_len member has not yet been set.
2518 */
2519 switch (ip->ip_v) {
2520#ifdef INET
2521 case IPVERSION:
2522 ipovly = (struct ipovly *)ip;
2523 ippseudo.ippseudo_src = ipovly->ih_src;
2524 ippseudo.ippseudo_dst = ipovly->ih_dst;
2525 ippseudo.ippseudo_pad = 0;
2526 ippseudo.ippseudo_p = IPPROTO_TCP;
2527 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2528 optlen);
2529 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2530
2531 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2532 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2533 break;
2534#endif
2535#ifdef INET6
2536 /*
2537 * RFC 2385, 2.0 Proposal
2538 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2539 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2540 * extended next header value (to form 32 bits), and 32-bit segment
2541 * length.
2542 * Note: Upper-Layer Packet Length comes before Next Header.
2543 */
2544 case (IPV6_VERSION >> 4):
2545 in6 = ip6->ip6_src;
2546 in6_clearscope(&in6);
2547 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2548 in6 = ip6->ip6_dst;
2549 in6_clearscope(&in6);
2550 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2551 plen = htonl(len + sizeof(struct tcphdr) + optlen);
2552 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2553 nhdr = 0;
2554 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2555 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2556 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2557 nhdr = IPPROTO_TCP;
2558 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2559
2560 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2561 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2562 break;
2563#endif
2564 default:
2565 KEY_FREESAV(&sav);
2566 return (-1);
2567 /* NOTREACHED */
2568 break;
2569 }
2570
2571
2572 /*
2573 * Step 2: Update MD5 hash with TCP header, excluding options.
2574 * The TCP checksum must be set to zero.
2575 */
2576 savecsum = th->th_sum;
2577 th->th_sum = 0;
2578 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2579 th->th_sum = savecsum;
2580
2581 /*
2582 * Step 3: Update MD5 hash with TCP segment data.
2583 * Use m_apply() to avoid an early m_pullup().
2584 */
2585 if (len > 0)
2586 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2587
2588 /*
2589 * Step 4: Update MD5 hash with shared secret.
2590 */
2591 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2592 MD5Final(buf, &ctx);
2593
2594 key_sa_recordxfer(sav, m);
2595 KEY_FREESAV(&sav);
2596 return (0);
2597}
2598
2599/*
2600 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2601 *
2602 * Return 0 if successful, otherwise return -1.
2603 */
2604int
2605tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
2606 u_char *buf, u_int direction)
2607{
2608 struct secasvar *sav;
2609
2610 if ((sav = tcp_get_sav(m, direction)) == NULL)
2611 return (-1);
2612
2613 return (tcp_signature_do_compute(m, len, optlen, buf, sav));
2614}
2615
2616/*
2617 * Verify the TCP-MD5 hash of a TCP segment. (RFC2385)
2618 *
2619 * Parameters:
2620 * m pointer to head of mbuf chain
2621 * len length of TCP segment data, excluding options
2622 * optlen length of TCP segment options
2623 * buf pointer to storage for computed MD5 digest
2624 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2625 *
2626 * Return 1 if successful, otherwise return 0.
2627 */
2628int
2629tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen,
2630 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
2631{
2632 char tmpdigest[TCP_SIGLEN];
2633
2634 if (tcp_sig_checksigs == 0)
2635 return (1);
2636 if ((tcpbflag & TF_SIGNATURE) == 0) {
2637 if ((to->to_flags & TOF_SIGNATURE) != 0) {
2638
2639 /*
2640 * If this socket is not expecting signature but
2641 * the segment contains signature just fail.
2642 */
2643 TCPSTAT_INC(tcps_sig_err_sigopt);
2644 TCPSTAT_INC(tcps_sig_rcvbadsig);
2645 return (0);
2646 }
2647
2648 /* Signature is not expected, and not present in segment. */
2649 return (1);
2650 }
2651
2652 /*
2653 * If this socket is expecting signature but the segment does not
2654 * contain any just fail.
2655 */
2656 if ((to->to_flags & TOF_SIGNATURE) == 0) {
2657 TCPSTAT_INC(tcps_sig_err_nosigopt);
2658 TCPSTAT_INC(tcps_sig_rcvbadsig);
2659 return (0);
2660 }
2661 if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0],
2662 IPSEC_DIR_INBOUND) == -1) {
2663 TCPSTAT_INC(tcps_sig_err_buildsig);
2664 TCPSTAT_INC(tcps_sig_rcvbadsig);
2665 return (0);
2666 }
2667
2668 if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) {
2669 TCPSTAT_INC(tcps_sig_rcvbadsig);
2670 return (0);
2671 }
2672 TCPSTAT_INC(tcps_sig_rcvgoodsig);
2673 return (1);
2674}
2675#endif /* TCP_SIGNATURE */
2676
2677static int
2678sysctl_drop(SYSCTL_HANDLER_ARGS)
2679{
2680 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2681 struct sockaddr_storage addrs[2];
2682 struct inpcb *inp;
2683 struct tcpcb *tp;
2684 struct tcptw *tw;
2685 struct sockaddr_in *fin, *lin;
2686#ifdef INET6
2687 struct sockaddr_in6 *fin6, *lin6;
2688#endif
2689 int error;
2690
2691 inp = NULL;
2692 fin = lin = NULL;
2693#ifdef INET6
2694 fin6 = lin6 = NULL;
2695#endif
2696 error = 0;
2697
2698 if (req->oldptr != NULL || req->oldlen != 0)
2699 return (EINVAL);
2700 if (req->newptr == NULL)
2701 return (EPERM);
2702 if (req->newlen < sizeof(addrs))
2703 return (ENOMEM);
2704 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2705 if (error)
2706 return (error);
2707
2708 switch (addrs[0].ss_family) {
2709#ifdef INET6
2710 case AF_INET6:
2711 fin6 = (struct sockaddr_in6 *)&addrs[0];
2712 lin6 = (struct sockaddr_in6 *)&addrs[1];
2713 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2714 lin6->sin6_len != sizeof(struct sockaddr_in6))
2715 return (EINVAL);
2716 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2717 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2718 return (EINVAL);
2719 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2720 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2721 fin = (struct sockaddr_in *)&addrs[0];
2722 lin = (struct sockaddr_in *)&addrs[1];
2723 break;
2724 }
2725 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2726 if (error)
2727 return (error);
2728 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2729 if (error)
2730 return (error);
2731 break;
2732#endif
2733#ifdef INET
2734 case AF_INET:
2735 fin = (struct sockaddr_in *)&addrs[0];
2736 lin = (struct sockaddr_in *)&addrs[1];
2737 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2738 lin->sin_len != sizeof(struct sockaddr_in))
2739 return (EINVAL);
2740 break;
2741#endif
2742 default:
2743 return (EINVAL);
2744 }
2745 INP_INFO_RLOCK(&V_tcbinfo);
2746 switch (addrs[0].ss_family) {
2747#ifdef INET6
2748 case AF_INET6:
2749 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2750 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2751 INPLOOKUP_WLOCKPCB, NULL);
2752 break;
2753#endif
2754#ifdef INET
2755 case AF_INET:
2756 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2757 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2758 break;
2759#endif
2760 }
2761 if (inp != NULL) {
2762 if (inp->inp_flags & INP_TIMEWAIT) {
2763 /*
2764 * XXXRW: There currently exists a state where an
2765 * inpcb is present, but its timewait state has been
2766 * discarded. For now, don't allow dropping of this
2767 * type of inpcb.
2768 */
2769 tw = intotw(inp);
2770 if (tw != NULL)
2771 tcp_twclose(tw, 0);
2772 else
2773 INP_WUNLOCK(inp);
2774 } else if (!(inp->inp_flags & INP_DROPPED) &&
2775 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2776 tp = intotcpcb(inp);
2777 tp = tcp_drop(tp, ECONNABORTED);
2778 if (tp != NULL)
2779 INP_WUNLOCK(inp);
2780 } else
2781 INP_WUNLOCK(inp);
2782 } else
2783 error = ESRCH;
2784 INP_INFO_RUNLOCK(&V_tcbinfo);
2785 return (error);
2786}
2787
2788SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2789 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
2790 0, sysctl_drop, "", "Drop TCP connection");
2791
2792/*
2793 * Generate a standardized TCP log line for use throughout the
2794 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2795 * allow use in the interrupt context.
2796 *
2797 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2798 * NB: The function may return NULL if memory allocation failed.
2799 *
2800 * Due to header inclusion and ordering limitations the struct ip
2801 * and ip6_hdr pointers have to be passed as void pointers.
2802 */
2803char *
2804tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2805 const void *ip6hdr)
2806{
2807
2808 /* Is logging enabled? */
2809 if (tcp_log_in_vain == 0)
2810 return (NULL);
2811
2812 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2813}
2814
2815char *
2816tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2817 const void *ip6hdr)
2818{
2819
2820 /* Is logging enabled? */
2821 if (tcp_log_debug == 0)
2822 return (NULL);
2823
2824 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2825}
2826
2827static char *
2828tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2829 const void *ip6hdr)
2830{
2831 char *s, *sp;
2832 size_t size;
2833 struct ip *ip;
2834#ifdef INET6
2835 const struct ip6_hdr *ip6;
2836
2837 ip6 = (const struct ip6_hdr *)ip6hdr;
2838#endif /* INET6 */
2839 ip = (struct ip *)ip4hdr;
2840
2841 /*
2842 * The log line looks like this:
2843 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2844 */
2845 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2846 sizeof(PRINT_TH_FLAGS) + 1 +
2847#ifdef INET6
2848 2 * INET6_ADDRSTRLEN;
2849#else
2850 2 * INET_ADDRSTRLEN;
2851#endif /* INET6 */
2852
2853 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2854 if (s == NULL)
2855 return (NULL);
2856
2857 strcat(s, "TCP: [");
2858 sp = s + strlen(s);
2859
2860 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2861 inet_ntoa_r(inc->inc_faddr, sp);
2862 sp = s + strlen(s);
2863 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2864 sp = s + strlen(s);
2865 inet_ntoa_r(inc->inc_laddr, sp);
2866 sp = s + strlen(s);
2867 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2868#ifdef INET6
2869 } else if (inc) {
2870 ip6_sprintf(sp, &inc->inc6_faddr);
2871 sp = s + strlen(s);
2872 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2873 sp = s + strlen(s);
2874 ip6_sprintf(sp, &inc->inc6_laddr);
2875 sp = s + strlen(s);
2876 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2877 } else if (ip6 && th) {
2878 ip6_sprintf(sp, &ip6->ip6_src);
2879 sp = s + strlen(s);
2880 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2881 sp = s + strlen(s);
2882 ip6_sprintf(sp, &ip6->ip6_dst);
2883 sp = s + strlen(s);
2884 sprintf(sp, "]:%i", ntohs(th->th_dport));
2885#endif /* INET6 */
2886#ifdef INET
2887 } else if (ip && th) {
2888 inet_ntoa_r(ip->ip_src, sp);
2889 sp = s + strlen(s);
2890 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2891 sp = s + strlen(s);
2892 inet_ntoa_r(ip->ip_dst, sp);
2893 sp = s + strlen(s);
2894 sprintf(sp, "]:%i", ntohs(th->th_dport));
2895#endif /* INET */
2896 } else {
2897 free(s, M_TCPLOG);
2898 return (NULL);
2899 }
2900 sp = s + strlen(s);
2901 if (th)
2902 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2903 if (*(s + size - 1) != '\0')
2904 panic("%s: string too long", __func__);
2905 return (s);
2906}
2907
2908/*
2909 * A subroutine which makes it easy to track TCP state changes with DTrace.
2910 * This function shouldn't be called for t_state initializations that don't
2911 * correspond to actual TCP state transitions.
2912 */
2913void
2914tcp_state_change(struct tcpcb *tp, int newstate)
2915{
2916#if defined(KDTRACE_HOOKS)
2917 int pstate = tp->t_state;
2918#endif
2919
2920 tp->t_state = newstate;
2921 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
2922}
|