1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
5 * The Regents of the University of California. All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * SUCH DAMAGE.
30 */
31
32 #include "opt_inet.h"
33 #include "opt_inet6.h"
34 #include "opt_ipsec.h"
35 #include "opt_kern_tls.h"
36
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/arb.h>
40 #include <sys/callout.h>
41 #include <sys/eventhandler.h>
42 #ifdef TCP_HHOOK
43 #include <sys/hhook.h>
44 #endif
45 #include <sys/kernel.h>
46 #ifdef TCP_HHOOK
47 #include <sys/khelp.h>
48 #endif
49 #ifdef KERN_TLS
50 #include <sys/ktls.h>
51 #endif
52 #include <sys/qmath.h>
53 #include <sys/stats.h>
54 #include <sys/sysctl.h>
55 #include <sys/jail.h>
56 #include <sys/malloc.h>
57 #include <sys/refcount.h>
58 #include <sys/mbuf.h>
59 #include <sys/priv.h>
60 #include <sys/sdt.h>
61 #include <sys/socket.h>
62 #include <sys/socketvar.h>
63 #include <sys/protosw.h>
64 #include <sys/random.h>
65
66 #include <vm/uma.h>
67
68 #include <net/route.h>
69 #include <net/route/nhop.h>
70 #include <net/if.h>
71 #include <net/if_var.h>
72 #include <net/if_private.h>
73 #include <net/vnet.h>
74
75 #include <netinet/in.h>
76 #include <netinet/in_fib.h>
77 #include <netinet/in_kdtrace.h>
78 #include <netinet/in_pcb.h>
79 #include <netinet/in_systm.h>
80 #include <netinet/in_var.h>
81 #include <netinet/ip.h>
82 #include <netinet/ip_icmp.h>
83 #include <netinet/ip_var.h>
84 #include <netinet/icmp_var.h>
85 #ifdef INET6
86 #include <netinet/icmp6.h>
87 #include <netinet/ip6.h>
88 #include <netinet6/in6_fib.h>
89 #include <netinet6/in6_pcb.h>
90 #include <netinet6/ip6_var.h>
91 #include <netinet6/scope6_var.h>
92 #include <netinet6/nd6.h>
93 #endif
94
95 #include <netinet/tcp.h>
96 #ifdef INVARIANTS
97 #define TCPSTATES
98 #endif
99 #include <netinet/tcp_fsm.h>
100 #include <netinet/tcp_seq.h>
101 #include <netinet/tcp_timer.h>
102 #include <netinet/tcp_var.h>
103 #include <netinet/tcp_ecn.h>
104 #include <netinet/tcp_log_buf.h>
105 #include <netinet/tcp_syncache.h>
106 #include <netinet/tcp_hpts.h>
107 #include <netinet/tcp_lro.h>
108 #include <netinet/cc/cc.h>
109 #include <netinet/tcpip.h>
110 #include <netinet/tcp_fastopen.h>
111 #include <netinet/tcp_accounting.h>
112 #ifdef TCP_OFFLOAD
113 #include <netinet/tcp_offload.h>
114 #endif
115 #include <netinet/udp.h>
116 #include <netinet/udp_var.h>
117 #ifdef INET6
118 #include <netinet6/tcp6_var.h>
119 #endif
120
121 #include <netipsec/ipsec_support.h>
122
123 #include <machine/in_cksum.h>
124 #include <crypto/siphash/siphash.h>
125
126 #include <security/mac/mac_framework.h>
127
128 #ifdef INET6
129 static ip6proto_ctlinput_t tcp6_ctlinput;
130 static udp_tun_icmp_t tcp6_ctlinput_viaudp;
131 #endif
132
133 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
134 #ifdef INET6
135 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
136 #endif
137
138 VNET_DEFINE(uint32_t, tcp_ack_war_time_window) = 1000;
139 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_timewindow,
140 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_ack_war_time_window), 0,
141 "Time interval in ms used to limit the number (ack_war_cnt) of challenge ACKs sent per TCP connection");
142 VNET_DEFINE(uint32_t, tcp_ack_war_cnt) = 5;
143 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_cnt, CTLFLAG_VNET | CTLFLAG_RW,
144 &VNET_NAME(tcp_ack_war_cnt), 0,
145 "Maximum number of challenge ACKs sent per TCP connection during the time interval (ack_war_timewindow)");
146
147 struct rwlock tcp_function_lock;
148
149 static int
sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)150 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
151 {
152 int error, new;
153
154 new = V_tcp_mssdflt;
155 error = sysctl_handle_int(oidp, &new, 0, req);
156 if (error == 0 && req->newptr) {
157 if (new < TCP_MINMSS)
158 error = EINVAL;
159 else
160 V_tcp_mssdflt = new;
161 }
162 return (error);
163 }
164
165 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
166 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
167 &VNET_NAME(tcp_mssdflt), 0, &sysctl_net_inet_tcp_mss_check, "I",
168 "Default TCP Maximum Segment Size");
169
170 #ifdef INET6
171 static int
sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)172 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
173 {
174 int error, new;
175
176 new = V_tcp_v6mssdflt;
177 error = sysctl_handle_int(oidp, &new, 0, req);
178 if (error == 0 && req->newptr) {
179 if (new < TCP_MINMSS)
180 error = EINVAL;
181 else
182 V_tcp_v6mssdflt = new;
183 }
184 return (error);
185 }
186
187 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
188 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
189 &VNET_NAME(tcp_v6mssdflt), 0, &sysctl_net_inet_tcp_mss_v6_check, "I",
190 "Default TCP Maximum Segment Size for IPv6");
191 #endif /* INET6 */
192
193 /*
194 * Minimum MSS we accept and use. This prevents DoS attacks where
195 * we are forced to a ridiculous low MSS like 20 and send hundreds
196 * of packets instead of one. The effect scales with the available
197 * bandwidth and quickly saturates the CPU and network interface
198 * with packet generation and sending. Set to zero to disable MINMSS
199 * checking. This setting prevents us from sending too small packets.
200 */
201 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
202 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
203 &VNET_NAME(tcp_minmss), 0,
204 "Minimum TCP Maximum Segment Size");
205
206 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
207 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
208 &VNET_NAME(tcp_do_rfc1323), 0,
209 "Enable rfc1323 (high performance TCP) extensions");
210
211 /*
212 * As of June 2021, several TCP stacks violate RFC 7323 from September 2014.
213 * Some stacks negotiate TS, but never send them after connection setup. Some
214 * stacks negotiate TS, but don't send them when sending keep-alive segments.
215 * These include modern widely deployed TCP stacks.
216 * Therefore tolerating violations for now...
217 */
218 VNET_DEFINE(int, tcp_tolerate_missing_ts) = 1;
219 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tolerate_missing_ts, CTLFLAG_VNET | CTLFLAG_RW,
220 &VNET_NAME(tcp_tolerate_missing_ts), 0,
221 "Tolerate missing TCP timestamps");
222
223 VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1;
224 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW,
225 &VNET_NAME(tcp_ts_offset_per_conn), 0,
226 "Initialize TCP timestamps per connection instead of per host pair");
227
228 /* How many connections are pacing */
229 static volatile uint32_t number_of_tcp_connections_pacing = 0;
230 static uint32_t shadow_num_connections = 0;
231 static counter_u64_t tcp_pacing_failures;
232 static counter_u64_t tcp_dgp_failures;
233 static uint32_t shadow_tcp_pacing_dgp = 0;
234 static volatile uint32_t number_of_dgp_connections = 0;
235
236 static int tcp_pacing_limit = 10000;
237 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pacing_limit, CTLFLAG_RW,
238 &tcp_pacing_limit, 1000,
239 "If the TCP stack does pacing, is there a limit (-1 = no, 0 = no pacing N = number of connections)");
240
241 static int tcp_dgp_limit = -1;
242 SYSCTL_INT(_net_inet_tcp, OID_AUTO, dgp_limit, CTLFLAG_RW,
243 &tcp_dgp_limit, -1,
244 "If the TCP stack does DGP, is there a limit (-1 = no, 0 = no dgp N = number of connections)");
245
246 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pacing_count, CTLFLAG_RD,
247 &shadow_num_connections, 0, "Number of TCP connections being paced");
248
249 SYSCTL_COUNTER_U64(_net_inet_tcp, OID_AUTO, pacing_failures, CTLFLAG_RD,
250 &tcp_pacing_failures, "Number of times we failed to enable pacing to avoid exceeding the limit");
251
252 SYSCTL_COUNTER_U64(_net_inet_tcp, OID_AUTO, dgp_failures, CTLFLAG_RD,
253 &tcp_dgp_failures, "Number of times we failed to enable dgp to avoid exceeding the limit");
254
255 static int tcp_log_debug = 0;
256 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
257 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
258
259 /*
260 * Target size of TCP PCB hash tables. Must be a power of two.
261 *
262 * Note that this can be overridden by the kernel environment
263 * variable net.inet.tcp.tcbhashsize
264 */
265 #ifndef TCBHASHSIZE
266 #define TCBHASHSIZE 0
267 #endif
268 static int tcp_tcbhashsize = TCBHASHSIZE;
269 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
270 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
271
272 static int do_tcpdrain = 1;
273 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
274 "Enable tcp_drain routine for extra help when low on mbufs");
275
276 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
277 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
278
279 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1;
280 #define V_icmp_may_rst VNET(icmp_may_rst)
281 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
282 &VNET_NAME(icmp_may_rst), 0,
283 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
284
285 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0;
286 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
287 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
288 &VNET_NAME(tcp_isn_reseed_interval), 0,
289 "Seconds between reseeding of ISN secret");
290
291 static int tcp_soreceive_stream;
292 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
293 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
294
295 VNET_DEFINE(uma_zone_t, sack_hole_zone);
296 #define V_sack_hole_zone VNET(sack_hole_zone)
297 VNET_DEFINE(uint32_t, tcp_map_entries_limit) = 0; /* unlimited */
298 static int
sysctl_net_inet_tcp_map_limit_check(SYSCTL_HANDLER_ARGS)299 sysctl_net_inet_tcp_map_limit_check(SYSCTL_HANDLER_ARGS)
300 {
301 int error;
302 uint32_t new;
303
304 new = V_tcp_map_entries_limit;
305 error = sysctl_handle_int(oidp, &new, 0, req);
306 if (error == 0 && req->newptr) {
307 /* only allow "0" and value > minimum */
308 if (new > 0 && new < TCP_MIN_MAP_ENTRIES_LIMIT)
309 error = EINVAL;
310 else
311 V_tcp_map_entries_limit = new;
312 }
313 return (error);
314 }
315 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, map_limit,
316 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
317 &VNET_NAME(tcp_map_entries_limit), 0,
318 &sysctl_net_inet_tcp_map_limit_check, "IU",
319 "Total sendmap entries limit");
320
321 VNET_DEFINE(uint32_t, tcp_map_split_limit) = 0; /* unlimited */
322 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, split_limit, CTLFLAG_VNET | CTLFLAG_RW,
323 &VNET_NAME(tcp_map_split_limit), 0,
324 "Total sendmap split entries limit");
325
326 #ifdef TCP_HHOOK
327 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
328 #endif
329
330 #define TS_OFFSET_SECRET_LENGTH SIPHASH_KEY_LENGTH
331 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]);
332 #define V_ts_offset_secret VNET(ts_offset_secret)
333
334 static int tcp_default_fb_init(struct tcpcb *tp, void **ptr);
335 static void tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged);
336 static int tcp_default_handoff_ok(struct tcpcb *tp);
337 static struct inpcb *tcp_notify(struct inpcb *, int);
338 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
339 static struct inpcb *tcp_mtudisc(struct inpcb *, int);
340 static struct inpcb *tcp_drop_syn_sent(struct inpcb *, int);
341 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
342 const void *ip4hdr, const void *ip6hdr);
343 static void tcp_default_switch_failed(struct tcpcb *tp);
344
345 #ifdef INET
346 static ipproto_ctlinput_t tcp_ctlinput;
347 static udp_tun_icmp_t tcp_ctlinput_viaudp;
348 #endif
349
350 static struct tcp_function_block tcp_def_funcblk = {
351 .tfb_tcp_block_name = "freebsd",
352 .tfb_tcp_output = tcp_default_output,
353 .tfb_tcp_do_segment = tcp_do_segment,
354 .tfb_tcp_ctloutput = tcp_default_ctloutput,
355 .tfb_tcp_handoff_ok = tcp_default_handoff_ok,
356 .tfb_tcp_fb_init = tcp_default_fb_init,
357 .tfb_tcp_fb_fini = tcp_default_fb_fini,
358 .tfb_switch_failed = tcp_default_switch_failed,
359 .tfb_flags = TCP_FUNC_DEFAULT_OK,
360 };
361
362 static int tcp_fb_cnt = 0;
363 struct tcp_funchead t_functions;
364 VNET_DEFINE_STATIC(struct tcp_function_block *, tcp_func_set_ptr) = &tcp_def_funcblk;
365 #define V_tcp_func_set_ptr VNET(tcp_func_set_ptr)
366
367 void
tcp_record_dsack(struct tcpcb * tp,tcp_seq start,tcp_seq end,int tlp)368 tcp_record_dsack(struct tcpcb *tp, tcp_seq start, tcp_seq end, int tlp)
369 {
370 TCPSTAT_INC(tcps_dsack_count);
371 tp->t_dsack_pack++;
372 if (tlp == 0) {
373 if (SEQ_GT(end, start)) {
374 tp->t_dsack_bytes += (end - start);
375 TCPSTAT_ADD(tcps_dsack_bytes, (end - start));
376 } else {
377 tp->t_dsack_tlp_bytes += (start - end);
378 TCPSTAT_ADD(tcps_dsack_bytes, (start - end));
379 }
380 } else {
381 if (SEQ_GT(end, start)) {
382 tp->t_dsack_bytes += (end - start);
383 TCPSTAT_ADD(tcps_dsack_tlp_bytes, (end - start));
384 } else {
385 tp->t_dsack_tlp_bytes += (start - end);
386 TCPSTAT_ADD(tcps_dsack_tlp_bytes, (start - end));
387 }
388 }
389 }
390
391 static struct tcp_function_block *
find_tcp_functions_locked(struct tcp_function_set * fs)392 find_tcp_functions_locked(struct tcp_function_set *fs)
393 {
394 struct tcp_function *f;
395 struct tcp_function_block *blk = NULL;
396
397 rw_assert(&tcp_function_lock, RA_LOCKED);
398 TAILQ_FOREACH(f, &t_functions, tf_next) {
399 if (strcmp(f->tf_name, fs->function_set_name) == 0) {
400 blk = f->tf_fb;
401 break;
402 }
403 }
404 return (blk);
405 }
406
407 static struct tcp_function_block *
find_tcp_fb_locked(struct tcp_function_block * blk,struct tcp_function ** s)408 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
409 {
410 struct tcp_function_block *rblk = NULL;
411 struct tcp_function *f;
412
413 rw_assert(&tcp_function_lock, RA_LOCKED);
414 TAILQ_FOREACH(f, &t_functions, tf_next) {
415 if (f->tf_fb == blk) {
416 rblk = blk;
417 if (s) {
418 *s = f;
419 }
420 break;
421 }
422 }
423 return (rblk);
424 }
425
426 struct tcp_function_block *
find_and_ref_tcp_functions(struct tcp_function_set * fs)427 find_and_ref_tcp_functions(struct tcp_function_set *fs)
428 {
429 struct tcp_function_block *blk;
430
431 rw_rlock(&tcp_function_lock);
432 blk = find_tcp_functions_locked(fs);
433 if (blk)
434 refcount_acquire(&blk->tfb_refcnt);
435 rw_runlock(&tcp_function_lock);
436 return (blk);
437 }
438
439 struct tcp_function_block *
find_and_ref_tcp_fb(struct tcp_function_block * blk)440 find_and_ref_tcp_fb(struct tcp_function_block *blk)
441 {
442 struct tcp_function_block *rblk;
443
444 rw_rlock(&tcp_function_lock);
445 rblk = find_tcp_fb_locked(blk, NULL);
446 if (rblk)
447 refcount_acquire(&rblk->tfb_refcnt);
448 rw_runlock(&tcp_function_lock);
449 return (rblk);
450 }
451
452 /* Find a matching alias for the given tcp_function_block. */
453 int
find_tcp_function_alias(struct tcp_function_block * blk,struct tcp_function_set * fs)454 find_tcp_function_alias(struct tcp_function_block *blk,
455 struct tcp_function_set *fs)
456 {
457 struct tcp_function *f;
458 int found;
459
460 found = 0;
461 rw_rlock(&tcp_function_lock);
462 TAILQ_FOREACH(f, &t_functions, tf_next) {
463 if ((f->tf_fb == blk) &&
464 (strncmp(f->tf_name, blk->tfb_tcp_block_name,
465 TCP_FUNCTION_NAME_LEN_MAX) != 0)) {
466 /* Matching function block with different name. */
467 strncpy(fs->function_set_name, f->tf_name,
468 TCP_FUNCTION_NAME_LEN_MAX);
469 found = 1;
470 break;
471 }
472 }
473 /* Null terminate the string appropriately. */
474 if (found) {
475 fs->function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
476 } else {
477 fs->function_set_name[0] = '\0';
478 }
479 rw_runlock(&tcp_function_lock);
480 return (found);
481 }
482
483 static struct tcp_function_block *
find_and_ref_tcp_default_fb(void)484 find_and_ref_tcp_default_fb(void)
485 {
486 struct tcp_function_block *rblk;
487
488 rw_rlock(&tcp_function_lock);
489 rblk = V_tcp_func_set_ptr;
490 refcount_acquire(&rblk->tfb_refcnt);
491 rw_runlock(&tcp_function_lock);
492 return (rblk);
493 }
494
495 void
tcp_switch_back_to_default(struct tcpcb * tp)496 tcp_switch_back_to_default(struct tcpcb *tp)
497 {
498 struct tcp_function_block *tfb;
499 void *ptr = NULL;
500
501 KASSERT(tp->t_fb != &tcp_def_funcblk,
502 ("%s: called by the built-in default stack", __func__));
503
504 if (tp->t_fb->tfb_tcp_timer_stop_all != NULL)
505 tp->t_fb->tfb_tcp_timer_stop_all(tp);
506
507 /*
508 * Now, we'll find a new function block to use.
509 * Start by trying the current user-selected
510 * default, unless this stack is the user-selected
511 * default.
512 */
513 tfb = find_and_ref_tcp_default_fb();
514 if (tfb == tp->t_fb) {
515 refcount_release(&tfb->tfb_refcnt);
516 tfb = NULL;
517 }
518 /* Does the stack accept this connection? */
519 if (tfb != NULL && (*tfb->tfb_tcp_handoff_ok)(tp)) {
520 refcount_release(&tfb->tfb_refcnt);
521 tfb = NULL;
522 }
523 /* Try to use that stack. */
524 if (tfb != NULL) {
525 /* Initialize the new stack. If it succeeds, we are done. */
526 if (tfb->tfb_tcp_fb_init == NULL ||
527 (*tfb->tfb_tcp_fb_init)(tp, &ptr) == 0) {
528 /* Release the old stack */
529 if (tp->t_fb->tfb_tcp_fb_fini != NULL)
530 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
531 refcount_release(&tp->t_fb->tfb_refcnt);
532 /* Now set in all the pointers */
533 tp->t_fb = tfb;
534 tp->t_fb_ptr = ptr;
535 return;
536 }
537 /*
538 * Initialization failed. Release the reference count on
539 * the looked up default stack.
540 */
541 refcount_release(&tfb->tfb_refcnt);
542 }
543
544 /*
545 * If that wasn't feasible, use the built-in default
546 * stack which is not allowed to reject anyone.
547 */
548 tfb = find_and_ref_tcp_fb(&tcp_def_funcblk);
549 if (tfb == NULL) {
550 /* there always should be a default */
551 panic("Can't refer to tcp_def_funcblk");
552 }
553 if ((*tfb->tfb_tcp_handoff_ok)(tp)) {
554 /* The default stack cannot say no */
555 panic("Default stack rejects a new session?");
556 }
557 if (tfb->tfb_tcp_fb_init != NULL &&
558 (*tfb->tfb_tcp_fb_init)(tp, &ptr)) {
559 /* The default stack cannot fail */
560 panic("Default stack initialization failed");
561 }
562 /* Now release the old stack */
563 if (tp->t_fb->tfb_tcp_fb_fini != NULL)
564 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
565 refcount_release(&tp->t_fb->tfb_refcnt);
566 /* And set in the pointers to the new */
567 tp->t_fb = tfb;
568 tp->t_fb_ptr = ptr;
569 }
570
571 static bool
tcp_recv_udp_tunneled_packet(struct mbuf * m,int off,struct inpcb * inp,const struct sockaddr * sa,void * ctx)572 tcp_recv_udp_tunneled_packet(struct mbuf *m, int off, struct inpcb *inp,
573 const struct sockaddr *sa, void *ctx)
574 {
575 struct ip *iph;
576 #ifdef INET6
577 struct ip6_hdr *ip6;
578 #endif
579 struct udphdr *uh;
580 struct tcphdr *th;
581 int len, thlen;
582 uint16_t port;
583
584 TCPSTAT_INC(tcps_tunneled_pkts);
585 if ((m->m_flags & M_PKTHDR) == 0) {
586 /* Can't handle one that is not a pkt hdr */
587 TCPSTAT_INC(tcps_tunneled_errs);
588 m_freem(m);
589 return (true);
590 }
591 thlen = sizeof(struct tcphdr);
592 if (m->m_len < off + sizeof(struct udphdr) + thlen &&
593 (m = m_pullup(m, off + sizeof(struct udphdr) + thlen)) == NULL) {
594 TCPSTAT_INC(tcps_tunneled_errs);
595 return (true);
596 }
597 iph = mtod(m, struct ip *);
598 uh = (struct udphdr *)((caddr_t)iph + off);
599 th = (struct tcphdr *)(uh + 1);
600 thlen = th->th_off << 2;
601 if (m->m_len < off + sizeof(struct udphdr) + thlen) {
602 m = m_pullup(m, off + sizeof(struct udphdr) + thlen);
603 if (m == NULL) {
604 TCPSTAT_INC(tcps_tunneled_errs);
605 return (true);
606 } else {
607 iph = mtod(m, struct ip *);
608 uh = (struct udphdr *)((caddr_t)iph + off);
609 th = (struct tcphdr *)(uh + 1);
610 }
611 }
612 m->m_pkthdr.tcp_tun_port = port = uh->uh_sport;
613 bcopy(th, uh, m->m_len - off - sizeof(struct udphdr));
614 m->m_len -= sizeof(struct udphdr);
615 m->m_pkthdr.len -= sizeof(struct udphdr);
616 /*
617 * We use the same algorithm for
618 * both UDP and TCP for c-sum. So
619 * the code in tcp_input will skip
620 * the checksum. So we do nothing
621 * with the flag (m->m_pkthdr.csum_flags).
622 */
623 switch (iph->ip_v) {
624 #ifdef INET
625 case IPVERSION:
626 len = ntohs(iph->ip_len) - sizeof(struct udphdr);
627 if (__predict_false(len != m->m_pkthdr.len)) {
628 TCPSTAT_INC(tcps_tunneled_errs);
629 m_freem(m);
630 return (true);
631 } else {
632 iph->ip_len = htons(len);
633 tcp_input_with_port(&m, &off, IPPROTO_TCP, port);
634 }
635 break;
636 #endif
637 #ifdef INET6
638 case IPV6_VERSION >> 4:
639 ip6 = mtod(m, struct ip6_hdr *);
640 len = ntohs(ip6->ip6_plen) - sizeof(struct udphdr);
641 if (__predict_false(len + sizeof(struct ip6_hdr) !=
642 m->m_pkthdr.len)) {
643 TCPSTAT_INC(tcps_tunneled_errs);
644 m_freem(m);
645 return (true);
646 } else {
647 ip6->ip6_plen = htons(len);
648 tcp6_input_with_port(&m, &off, IPPROTO_TCP, port);
649 }
650 break;
651 #endif
652 default:
653 m_freem(m);
654 break;
655 }
656 return (true);
657 }
658
659 static int
sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)660 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
661 {
662 struct tcp_function_set fs;
663 struct tcp_function_block *blk;
664 int error;
665
666 memset(&fs, 0, sizeof(struct tcp_function_set));
667 rw_rlock(&tcp_function_lock);
668 blk = find_tcp_fb_locked(V_tcp_func_set_ptr, NULL);
669 if (blk != NULL) {
670 /* Found him */
671 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
672 fs.pcbcnt = blk->tfb_refcnt;
673 }
674 rw_runlock(&tcp_function_lock);
675 error = sysctl_handle_string(oidp, fs.function_set_name,
676 sizeof(fs.function_set_name), req);
677
678 /* Check for error or no change */
679 if (error != 0 || req->newptr == NULL)
680 return (error);
681
682 rw_wlock(&tcp_function_lock);
683 blk = find_tcp_functions_locked(&fs);
684 if ((blk == NULL) ||
685 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
686 error = ENOENT;
687 goto done;
688 }
689 if ((blk->tfb_flags & TCP_FUNC_DEFAULT_OK) == 0) {
690 error = EINVAL;
691 goto done;
692 }
693 V_tcp_func_set_ptr = blk;
694 done:
695 rw_wunlock(&tcp_function_lock);
696 return (error);
697 }
698
699 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
700 CTLFLAG_VNET | CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
701 NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
702 "Set/get the default TCP functions");
703
704 static int
sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)705 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
706 {
707 int error, cnt, linesz;
708 struct tcp_function *f;
709 char *buffer, *cp;
710 size_t bufsz, outsz;
711 bool alias;
712
713 cnt = 0;
714 rw_rlock(&tcp_function_lock);
715 TAILQ_FOREACH(f, &t_functions, tf_next) {
716 cnt++;
717 }
718 rw_runlock(&tcp_function_lock);
719
720 bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
721 buffer = malloc(bufsz, M_TEMP, M_WAITOK);
722
723 error = 0;
724 cp = buffer;
725
726 linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
727 "Alias", "PCB count");
728 cp += linesz;
729 bufsz -= linesz;
730 outsz = linesz;
731
732 rw_rlock(&tcp_function_lock);
733 TAILQ_FOREACH(f, &t_functions, tf_next) {
734 alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
735 linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
736 f->tf_fb->tfb_tcp_block_name,
737 (f->tf_fb == V_tcp_func_set_ptr) ? '*' : ' ',
738 alias ? f->tf_name : "-",
739 f->tf_fb->tfb_refcnt);
740 if (linesz >= bufsz) {
741 error = EOVERFLOW;
742 break;
743 }
744 cp += linesz;
745 bufsz -= linesz;
746 outsz += linesz;
747 }
748 rw_runlock(&tcp_function_lock);
749 if (error == 0)
750 error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
751 free(buffer, M_TEMP);
752 return (error);
753 }
754
755 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
756 CTLFLAG_VNET | CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
757 NULL, 0, sysctl_net_inet_list_available, "A",
758 "list available TCP Function sets");
759
760 VNET_DEFINE(int, tcp_udp_tunneling_port) = TCP_TUNNELING_PORT_DEFAULT;
761
762 #ifdef INET
763 VNET_DEFINE(struct socket *, udp4_tun_socket) = NULL;
764 #define V_udp4_tun_socket VNET(udp4_tun_socket)
765 #endif
766 #ifdef INET6
767 VNET_DEFINE(struct socket *, udp6_tun_socket) = NULL;
768 #define V_udp6_tun_socket VNET(udp6_tun_socket)
769 #endif
770
771 static struct sx tcpoudp_lock;
772
773 static void
tcp_over_udp_stop(void)774 tcp_over_udp_stop(void)
775 {
776
777 sx_assert(&tcpoudp_lock, SA_XLOCKED);
778
779 #ifdef INET
780 if (V_udp4_tun_socket != NULL) {
781 soclose(V_udp4_tun_socket);
782 V_udp4_tun_socket = NULL;
783 }
784 #endif
785 #ifdef INET6
786 if (V_udp6_tun_socket != NULL) {
787 soclose(V_udp6_tun_socket);
788 V_udp6_tun_socket = NULL;
789 }
790 #endif
791 }
792
793 static int
tcp_over_udp_start(void)794 tcp_over_udp_start(void)
795 {
796 uint16_t port;
797 int ret;
798 #ifdef INET
799 struct sockaddr_in sin;
800 #endif
801 #ifdef INET6
802 struct sockaddr_in6 sin6;
803 #endif
804
805 sx_assert(&tcpoudp_lock, SA_XLOCKED);
806
807 port = V_tcp_udp_tunneling_port;
808 if (ntohs(port) == 0) {
809 /* Must have a port set */
810 return (EINVAL);
811 }
812 #ifdef INET
813 if (V_udp4_tun_socket != NULL) {
814 /* Already running -- must stop first */
815 return (EALREADY);
816 }
817 #endif
818 #ifdef INET6
819 if (V_udp6_tun_socket != NULL) {
820 /* Already running -- must stop first */
821 return (EALREADY);
822 }
823 #endif
824 #ifdef INET
825 if ((ret = socreate(PF_INET, &V_udp4_tun_socket,
826 SOCK_DGRAM, IPPROTO_UDP,
827 curthread->td_ucred, curthread))) {
828 tcp_over_udp_stop();
829 return (ret);
830 }
831 /* Call the special UDP hook. */
832 if ((ret = udp_set_kernel_tunneling(V_udp4_tun_socket,
833 tcp_recv_udp_tunneled_packet,
834 tcp_ctlinput_viaudp,
835 NULL))) {
836 tcp_over_udp_stop();
837 return (ret);
838 }
839 /* Ok, we have a socket, bind it to the port. */
840 memset(&sin, 0, sizeof(struct sockaddr_in));
841 sin.sin_len = sizeof(struct sockaddr_in);
842 sin.sin_family = AF_INET;
843 sin.sin_port = htons(port);
844 if ((ret = sobind(V_udp4_tun_socket,
845 (struct sockaddr *)&sin, curthread))) {
846 tcp_over_udp_stop();
847 return (ret);
848 }
849 #endif
850 #ifdef INET6
851 if ((ret = socreate(PF_INET6, &V_udp6_tun_socket,
852 SOCK_DGRAM, IPPROTO_UDP,
853 curthread->td_ucred, curthread))) {
854 tcp_over_udp_stop();
855 return (ret);
856 }
857 /* Call the special UDP hook. */
858 if ((ret = udp_set_kernel_tunneling(V_udp6_tun_socket,
859 tcp_recv_udp_tunneled_packet,
860 tcp6_ctlinput_viaudp,
861 NULL))) {
862 tcp_over_udp_stop();
863 return (ret);
864 }
865 /* Ok, we have a socket, bind it to the port. */
866 memset(&sin6, 0, sizeof(struct sockaddr_in6));
867 sin6.sin6_len = sizeof(struct sockaddr_in6);
868 sin6.sin6_family = AF_INET6;
869 sin6.sin6_port = htons(port);
870 if ((ret = sobind(V_udp6_tun_socket,
871 (struct sockaddr *)&sin6, curthread))) {
872 tcp_over_udp_stop();
873 return (ret);
874 }
875 #endif
876 return (0);
877 }
878
879 static int
sysctl_net_inet_tcp_udp_tunneling_port_check(SYSCTL_HANDLER_ARGS)880 sysctl_net_inet_tcp_udp_tunneling_port_check(SYSCTL_HANDLER_ARGS)
881 {
882 int error;
883 uint32_t old, new;
884
885 old = V_tcp_udp_tunneling_port;
886 new = old;
887 error = sysctl_handle_int(oidp, &new, 0, req);
888 if ((error == 0) &&
889 (req->newptr != NULL)) {
890 if ((new < TCP_TUNNELING_PORT_MIN) ||
891 (new > TCP_TUNNELING_PORT_MAX)) {
892 error = EINVAL;
893 } else {
894 sx_xlock(&tcpoudp_lock);
895 V_tcp_udp_tunneling_port = new;
896 if (old != 0) {
897 tcp_over_udp_stop();
898 }
899 if (new != 0) {
900 error = tcp_over_udp_start();
901 if (error != 0) {
902 V_tcp_udp_tunneling_port = 0;
903 }
904 }
905 sx_xunlock(&tcpoudp_lock);
906 }
907 }
908 return (error);
909 }
910
911 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_port,
912 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
913 &VNET_NAME(tcp_udp_tunneling_port),
914 0, &sysctl_net_inet_tcp_udp_tunneling_port_check, "IU",
915 "Tunneling port for tcp over udp");
916
917 VNET_DEFINE(int, tcp_udp_tunneling_overhead) = TCP_TUNNELING_OVERHEAD_DEFAULT;
918
919 static int
sysctl_net_inet_tcp_udp_tunneling_overhead_check(SYSCTL_HANDLER_ARGS)920 sysctl_net_inet_tcp_udp_tunneling_overhead_check(SYSCTL_HANDLER_ARGS)
921 {
922 int error, new;
923
924 new = V_tcp_udp_tunneling_overhead;
925 error = sysctl_handle_int(oidp, &new, 0, req);
926 if (error == 0 && req->newptr) {
927 if ((new < TCP_TUNNELING_OVERHEAD_MIN) ||
928 (new > TCP_TUNNELING_OVERHEAD_MAX))
929 error = EINVAL;
930 else
931 V_tcp_udp_tunneling_overhead = new;
932 }
933 return (error);
934 }
935
936 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_overhead,
937 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
938 &VNET_NAME(tcp_udp_tunneling_overhead),
939 0, &sysctl_net_inet_tcp_udp_tunneling_overhead_check, "IU",
940 "MSS reduction when using tcp over udp");
941
942 /*
943 * Exports one (struct tcp_function_info) for each alias/name.
944 */
945 static int
sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS)946 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS)
947 {
948 int cnt, error;
949 struct tcp_function *f;
950 struct tcp_function_info tfi;
951
952 /*
953 * We don't allow writes.
954 */
955 if (req->newptr != NULL)
956 return (EINVAL);
957
958 /*
959 * Wire the old buffer so we can directly copy the functions to
960 * user space without dropping the lock.
961 */
962 if (req->oldptr != NULL) {
963 error = sysctl_wire_old_buffer(req, 0);
964 if (error)
965 return (error);
966 }
967
968 /*
969 * Walk the list and copy out matching entries. If INVARIANTS
970 * is compiled in, also walk the list to verify the length of
971 * the list matches what we have recorded.
972 */
973 rw_rlock(&tcp_function_lock);
974
975 cnt = 0;
976 #ifndef INVARIANTS
977 if (req->oldptr == NULL) {
978 cnt = tcp_fb_cnt;
979 goto skip_loop;
980 }
981 #endif
982 TAILQ_FOREACH(f, &t_functions, tf_next) {
983 #ifdef INVARIANTS
984 cnt++;
985 #endif
986 if (req->oldptr != NULL) {
987 bzero(&tfi, sizeof(tfi));
988 tfi.tfi_refcnt = f->tf_fb->tfb_refcnt;
989 tfi.tfi_id = f->tf_fb->tfb_id;
990 (void)strlcpy(tfi.tfi_alias, f->tf_name,
991 sizeof(tfi.tfi_alias));
992 (void)strlcpy(tfi.tfi_name,
993 f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name));
994 error = SYSCTL_OUT(req, &tfi, sizeof(tfi));
995 /*
996 * Don't stop on error, as that is the
997 * mechanism we use to accumulate length
998 * information if the buffer was too short.
999 */
1000 }
1001 }
1002 KASSERT(cnt == tcp_fb_cnt,
1003 ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt));
1004 #ifndef INVARIANTS
1005 skip_loop:
1006 #endif
1007 rw_runlock(&tcp_function_lock);
1008 if (req->oldptr == NULL)
1009 error = SYSCTL_OUT(req, NULL,
1010 (cnt + 1) * sizeof(struct tcp_function_info));
1011
1012 return (error);
1013 }
1014
1015 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info,
1016 CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE,
1017 NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info",
1018 "List TCP function block name-to-ID mappings");
1019
1020 /*
1021 * tfb_tcp_handoff_ok() function for the default stack.
1022 * Note that we'll basically try to take all comers.
1023 */
1024 static int
tcp_default_handoff_ok(struct tcpcb * tp)1025 tcp_default_handoff_ok(struct tcpcb *tp)
1026 {
1027
1028 return (0);
1029 }
1030
1031 /*
1032 * tfb_tcp_fb_init() function for the default stack.
1033 *
1034 * This handles making sure we have appropriate timers set if you are
1035 * transitioning a socket that has some amount of setup done.
1036 *
1037 * The init() fuction from the default can *never* return non-zero i.e.
1038 * it is required to always succeed since it is the stack of last resort!
1039 */
1040 static int
tcp_default_fb_init(struct tcpcb * tp,void ** ptr)1041 tcp_default_fb_init(struct tcpcb *tp, void **ptr)
1042 {
1043 struct socket *so = tptosocket(tp);
1044 int rexmt;
1045
1046 INP_WLOCK_ASSERT(tptoinpcb(tp));
1047 /* We don't use the pointer */
1048 *ptr = NULL;
1049
1050 /* Make sure we get no interesting mbuf queuing behavior */
1051 /* All mbuf queue/ack compress flags should be off */
1052 tcp_lro_features_off(tp);
1053
1054 /* Cancel the GP measurement in progress */
1055 tp->t_flags &= ~TF_GPUTINPROG;
1056 /* Validate the timers are not in usec, if they are convert */
1057 tcp_change_time_units(tp, TCP_TMR_GRANULARITY_TICKS);
1058 if ((tp->t_state == TCPS_SYN_SENT) ||
1059 (tp->t_state == TCPS_SYN_RECEIVED))
1060 rexmt = tcp_rexmit_initial * tcp_backoff[tp->t_rxtshift];
1061 else
1062 rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
1063 if (tp->t_rxtshift == 0)
1064 tp->t_rxtcur = rexmt;
1065 else
1066 TCPT_RANGESET(tp->t_rxtcur, rexmt, tp->t_rttmin,
1067 tcp_rexmit_max);
1068
1069 /*
1070 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't
1071 * know what to do for unexpected states (which includes TIME_WAIT).
1072 */
1073 if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT)
1074 return (0);
1075
1076 /*
1077 * Make sure some kind of transmission timer is set if there is
1078 * outstanding data.
1079 */
1080 if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) ||
1081 tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) ||
1082 tcp_timer_active(tp, TT_PERSIST))) {
1083 /*
1084 * If the session has established and it looks like it should
1085 * be in the persist state, set the persist timer. Otherwise,
1086 * set the retransmit timer.
1087 */
1088 if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 &&
1089 (int32_t)(tp->snd_nxt - tp->snd_una) <
1090 (int32_t)sbavail(&so->so_snd))
1091 tcp_setpersist(tp);
1092 else
1093 tcp_timer_activate(tp, TT_REXMT, TP_RXTCUR(tp));
1094 }
1095
1096 /* All non-embryonic sessions get a keepalive timer. */
1097 if (!tcp_timer_active(tp, TT_KEEP))
1098 tcp_timer_activate(tp, TT_KEEP,
1099 TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) :
1100 TP_KEEPINIT(tp));
1101
1102 /*
1103 * Make sure critical variables are initialized
1104 * if transitioning while in Recovery.
1105 */
1106 if IN_FASTRECOVERY(tp->t_flags) {
1107 if (tp->sackhint.recover_fs == 0)
1108 tp->sackhint.recover_fs = max(1,
1109 tp->snd_nxt - tp->snd_una);
1110 }
1111
1112 return (0);
1113 }
1114
1115 /*
1116 * tfb_tcp_fb_fini() function for the default stack.
1117 *
1118 * This changes state as necessary (or prudent) to prepare for another stack
1119 * to assume responsibility for the connection.
1120 */
1121 static void
tcp_default_fb_fini(struct tcpcb * tp,int tcb_is_purged)1122 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged)
1123 {
1124
1125 INP_WLOCK_ASSERT(tptoinpcb(tp));
1126
1127 #ifdef TCP_BLACKBOX
1128 tcp_log_flowend(tp);
1129 #endif
1130 tp->t_acktime = 0;
1131 return;
1132 }
1133
1134 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
1135 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
1136
1137 static struct mtx isn_mtx;
1138
1139 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
1140 #define ISN_LOCK() mtx_lock(&isn_mtx)
1141 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
1142
1143 INPCBSTORAGE_DEFINE(tcpcbstor, tcpcb, "tcpinp", "tcp_inpcb", "tcp", "tcphash");
1144
1145 /*
1146 * Take a value and get the next power of 2 that doesn't overflow.
1147 * Used to size the tcp_inpcb hash buckets.
1148 */
1149 static int
maketcp_hashsize(int size)1150 maketcp_hashsize(int size)
1151 {
1152 int hashsize;
1153
1154 /*
1155 * auto tune.
1156 * get the next power of 2 higher than maxsockets.
1157 */
1158 hashsize = 1 << fls(size);
1159 /* catch overflow, and just go one power of 2 smaller */
1160 if (hashsize < size) {
1161 hashsize = 1 << (fls(size) - 1);
1162 }
1163 return (hashsize);
1164 }
1165
1166 static volatile int next_tcp_stack_id = 1;
1167
1168 /*
1169 * Register a TCP function block with the name provided in the names
1170 * array. (Note that this function does NOT automatically register
1171 * blk->tfb_tcp_block_name as a stack name. Therefore, you should
1172 * explicitly include blk->tfb_tcp_block_name in the list of names if
1173 * you wish to register the stack with that name.)
1174 *
1175 * Either all name registrations will succeed or all will fail. If
1176 * a name registration fails, the function will update the num_names
1177 * argument to point to the array index of the name that encountered
1178 * the failure.
1179 *
1180 * Returns 0 on success, or an error code on failure.
1181 */
1182 int
register_tcp_functions_as_names(struct tcp_function_block * blk,int wait,const char * names[],int * num_names)1183 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
1184 const char *names[], int *num_names)
1185 {
1186 struct tcp_function *f[TCP_FUNCTION_NAME_NUM_MAX];
1187 struct tcp_function_set fs;
1188 int error, i, num_registered;
1189
1190 KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
1191 KASSERT(*num_names > 0,
1192 ("%s: Called with non-positive length of name list", __func__));
1193 KASSERT(rw_initialized(&tcp_function_lock),
1194 ("%s: called too early", __func__));
1195
1196 if (*num_names > TCP_FUNCTION_NAME_NUM_MAX) {
1197 /* Too many names. */
1198 *num_names = 0;
1199 return (E2BIG);
1200 }
1201 if ((blk->tfb_tcp_output == NULL) ||
1202 (blk->tfb_tcp_do_segment == NULL) ||
1203 (blk->tfb_tcp_ctloutput == NULL) ||
1204 (blk->tfb_tcp_handoff_ok == NULL) ||
1205 (strlen(blk->tfb_tcp_block_name) == 0)) {
1206 /* These functions are required and a name is needed. */
1207 *num_names = 0;
1208 return (EINVAL);
1209 }
1210
1211 for (i = 0; i < *num_names; i++) {
1212 f[i] = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
1213 if (f[i] == NULL) {
1214 while (--i >= 0)
1215 free(f[i], M_TCPFUNCTIONS);
1216 *num_names = 0;
1217 return (ENOMEM);
1218 }
1219 }
1220
1221 num_registered = 0;
1222 rw_wlock(&tcp_function_lock);
1223 if (find_tcp_fb_locked(blk, NULL) != NULL) {
1224 /* A TCP function block can only be registered once. */
1225 error = EALREADY;
1226 goto cleanup;
1227 }
1228 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1229 error = EINVAL;
1230 goto cleanup;
1231 }
1232 refcount_init(&blk->tfb_refcnt, 0);
1233 blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1);
1234 for (i = 0; i < *num_names; i++) {
1235 (void)strlcpy(fs.function_set_name, names[i],
1236 sizeof(fs.function_set_name));
1237 if (find_tcp_functions_locked(&fs) != NULL) {
1238 /* Duplicate name space not allowed */
1239 error = EALREADY;
1240 goto cleanup;
1241 }
1242 f[i]->tf_fb = blk;
1243 (void)strlcpy(f[i]->tf_name, names[i], sizeof(f[i]->tf_name));
1244 TAILQ_INSERT_TAIL(&t_functions, f[i], tf_next);
1245 tcp_fb_cnt++;
1246 num_registered++;
1247 }
1248 rw_wunlock(&tcp_function_lock);
1249 return (0);
1250
1251 cleanup:
1252 /* Remove the entries just added. */
1253 for (i = 0; i < *num_names; i++) {
1254 if (i < num_registered) {
1255 TAILQ_REMOVE(&t_functions, f[i], tf_next);
1256 tcp_fb_cnt--;
1257 }
1258 f[i]->tf_fb = NULL;
1259 free(f[i], M_TCPFUNCTIONS);
1260 }
1261 rw_wunlock(&tcp_function_lock);
1262 *num_names = num_registered;
1263 return (error);
1264 }
1265
1266 /*
1267 * Register a TCP function block using the name provided in the name
1268 * argument.
1269 *
1270 * Returns 0 on success, or an error code on failure.
1271 */
1272 int
register_tcp_functions_as_name(struct tcp_function_block * blk,const char * name,int wait)1273 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
1274 int wait)
1275 {
1276 const char *name_list[1];
1277 int num_names, rv;
1278
1279 num_names = 1;
1280 if (name != NULL)
1281 name_list[0] = name;
1282 else
1283 name_list[0] = blk->tfb_tcp_block_name;
1284 rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
1285 return (rv);
1286 }
1287
1288 /*
1289 * Register a TCP function block using the name defined in
1290 * blk->tfb_tcp_block_name.
1291 *
1292 * Returns 0 on success, or an error code on failure.
1293 */
1294 int
register_tcp_functions(struct tcp_function_block * blk,int wait)1295 register_tcp_functions(struct tcp_function_block *blk, int wait)
1296 {
1297
1298 return (register_tcp_functions_as_name(blk, NULL, wait));
1299 }
1300
1301 /*
1302 * Deregister all names associated with a function block. This
1303 * functionally removes the function block from use within the system.
1304 *
1305 * When called with a true quiesce argument, mark the function block
1306 * as being removed so no more stacks will use it and determine
1307 * whether the removal would succeed.
1308 *
1309 * When called with a false quiesce argument, actually attempt the
1310 * removal.
1311 *
1312 * When called with a force argument, attempt to switch all TCBs to
1313 * use the default stack instead of returning EBUSY.
1314 *
1315 * Returns 0 on success (or if the removal would succeed), or an error
1316 * code on failure.
1317 */
1318 int
deregister_tcp_functions(struct tcp_function_block * blk,bool quiesce,bool force)1319 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce,
1320 bool force)
1321 {
1322 struct tcp_function *f;
1323 VNET_ITERATOR_DECL(vnet_iter);
1324
1325 if (blk == &tcp_def_funcblk) {
1326 /* You can't un-register the default */
1327 return (EPERM);
1328 }
1329 rw_wlock(&tcp_function_lock);
1330 VNET_LIST_RLOCK_NOSLEEP();
1331 VNET_FOREACH(vnet_iter) {
1332 CURVNET_SET(vnet_iter);
1333 if (blk == V_tcp_func_set_ptr) {
1334 /* You can't free the current default in some vnet. */
1335 CURVNET_RESTORE();
1336 VNET_LIST_RUNLOCK_NOSLEEP();
1337 rw_wunlock(&tcp_function_lock);
1338 return (EBUSY);
1339 }
1340 CURVNET_RESTORE();
1341 }
1342 VNET_LIST_RUNLOCK_NOSLEEP();
1343 /* Mark the block so no more stacks can use it. */
1344 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
1345 /*
1346 * If TCBs are still attached to the stack, attempt to switch them
1347 * to the default stack.
1348 */
1349 if (force && blk->tfb_refcnt) {
1350 struct inpcb *inp;
1351 struct tcpcb *tp;
1352 VNET_ITERATOR_DECL(vnet_iter);
1353
1354 rw_wunlock(&tcp_function_lock);
1355
1356 VNET_LIST_RLOCK();
1357 VNET_FOREACH(vnet_iter) {
1358 CURVNET_SET(vnet_iter);
1359 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
1360 INPLOOKUP_WLOCKPCB);
1361
1362 while ((inp = inp_next(&inpi)) != NULL) {
1363 tp = intotcpcb(inp);
1364 if (tp == NULL || tp->t_fb != blk)
1365 continue;
1366 tcp_switch_back_to_default(tp);
1367 }
1368 CURVNET_RESTORE();
1369 }
1370 VNET_LIST_RUNLOCK();
1371
1372 rw_wlock(&tcp_function_lock);
1373 }
1374 if (blk->tfb_refcnt) {
1375 /* TCBs still attached. */
1376 rw_wunlock(&tcp_function_lock);
1377 return (EBUSY);
1378 }
1379 if (quiesce) {
1380 /* Skip removal. */
1381 rw_wunlock(&tcp_function_lock);
1382 return (0);
1383 }
1384 /* Remove any function names that map to this function block. */
1385 while (find_tcp_fb_locked(blk, &f) != NULL) {
1386 TAILQ_REMOVE(&t_functions, f, tf_next);
1387 tcp_fb_cnt--;
1388 f->tf_fb = NULL;
1389 free(f, M_TCPFUNCTIONS);
1390 }
1391 rw_wunlock(&tcp_function_lock);
1392 return (0);
1393 }
1394
1395 static void
tcp_drain(void * ctx __unused,int flags __unused)1396 tcp_drain(void *ctx __unused, int flags __unused)
1397 {
1398 struct epoch_tracker et;
1399 VNET_ITERATOR_DECL(vnet_iter);
1400
1401 if (!do_tcpdrain)
1402 return;
1403
1404 NET_EPOCH_ENTER(et);
1405 VNET_LIST_RLOCK_NOSLEEP();
1406 VNET_FOREACH(vnet_iter) {
1407 CURVNET_SET(vnet_iter);
1408 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
1409 INPLOOKUP_WLOCKPCB);
1410 struct inpcb *inpb;
1411 struct tcpcb *tcpb;
1412
1413 /*
1414 * Walk the tcpbs, if existing, and flush the reassembly queue,
1415 * if there is one...
1416 * XXX: The "Net/3" implementation doesn't imply that the TCP
1417 * reassembly queue should be flushed, but in a situation
1418 * where we're really low on mbufs, this is potentially
1419 * useful.
1420 */
1421 while ((inpb = inp_next(&inpi)) != NULL) {
1422 if ((tcpb = intotcpcb(inpb)) != NULL) {
1423 tcp_reass_flush(tcpb);
1424 tcp_clean_sackreport(tcpb);
1425 #ifdef TCP_BLACKBOX
1426 tcp_log_drain(tcpb);
1427 #endif
1428 }
1429 }
1430 CURVNET_RESTORE();
1431 }
1432 VNET_LIST_RUNLOCK_NOSLEEP();
1433 NET_EPOCH_EXIT(et);
1434 }
1435
1436 static void
tcp_vnet_init(void * arg __unused)1437 tcp_vnet_init(void *arg __unused)
1438 {
1439
1440 #ifdef TCP_HHOOK
1441 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
1442 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1443 printf("%s: WARNING: unable to register helper hook\n", __func__);
1444 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
1445 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1446 printf("%s: WARNING: unable to register helper hook\n", __func__);
1447 #endif
1448 #ifdef STATS
1449 if (tcp_stats_init())
1450 printf("%s: WARNING: unable to initialise TCP stats\n",
1451 __func__);
1452 #endif
1453 in_pcbinfo_init(&V_tcbinfo, &tcpcbstor, tcp_tcbhashsize,
1454 tcp_tcbhashsize);
1455
1456 syncache_init();
1457 tcp_hc_init();
1458
1459 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
1460 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
1461 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1462
1463 tcp_fastopen_init();
1464
1465 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK);
1466 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK);
1467
1468 V_tcp_msl = TCPTV_MSL;
1469 V_tcp_msl_local = TCPTV_MSL_LOCAL;
1470 arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0);
1471 }
1472 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
1473 tcp_vnet_init, NULL);
1474
1475 static void
tcp_init(void * arg __unused)1476 tcp_init(void *arg __unused)
1477 {
1478 int hashsize;
1479
1480 tcp_reass_global_init();
1481
1482 /* XXX virtualize those below? */
1483 tcp_delacktime = TCPTV_DELACK;
1484 tcp_keepinit = TCPTV_KEEP_INIT;
1485 tcp_keepidle = TCPTV_KEEP_IDLE;
1486 tcp_keepintvl = TCPTV_KEEPINTVL;
1487 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
1488 tcp_rexmit_initial = TCPTV_RTOBASE;
1489 tcp_rexmit_min = TCPTV_MIN;
1490 tcp_rexmit_max = TCPTV_REXMTMAX;
1491 tcp_persmin = TCPTV_PERSMIN;
1492 tcp_persmax = TCPTV_PERSMAX;
1493 tcp_rexmit_slop = TCPTV_CPU_VAR;
1494 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
1495
1496 /* Setup the tcp function block list */
1497 TAILQ_INIT(&t_functions);
1498 rw_init(&tcp_function_lock, "tcp_func_lock");
1499 register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
1500 sx_init(&tcpoudp_lock, "TCP over UDP configuration");
1501 #ifdef TCP_BLACKBOX
1502 /* Initialize the TCP logging data. */
1503 tcp_log_init();
1504 #endif
1505
1506 if (tcp_soreceive_stream) {
1507 #ifdef INET
1508 tcp_protosw.pr_soreceive = soreceive_stream;
1509 #endif
1510 #ifdef INET6
1511 tcp6_protosw.pr_soreceive = soreceive_stream;
1512 #endif /* INET6 */
1513 }
1514
1515 #ifdef INET6
1516 max_protohdr_grow(sizeof(struct ip6_hdr) + sizeof(struct tcphdr));
1517 #else /* INET6 */
1518 max_protohdr_grow(sizeof(struct tcpiphdr));
1519 #endif /* INET6 */
1520
1521 ISN_LOCK_INIT();
1522 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
1523 SHUTDOWN_PRI_DEFAULT);
1524 EVENTHANDLER_REGISTER(vm_lowmem, tcp_drain, NULL, LOWMEM_PRI_DEFAULT);
1525 EVENTHANDLER_REGISTER(mbuf_lowmem, tcp_drain, NULL, LOWMEM_PRI_DEFAULT);
1526
1527 tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK);
1528 tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK);
1529 tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK);
1530 tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK);
1531 tcp_extra_mbuf = counter_u64_alloc(M_WAITOK);
1532 tcp_would_have_but = counter_u64_alloc(M_WAITOK);
1533 tcp_comp_total = counter_u64_alloc(M_WAITOK);
1534 tcp_uncomp_total = counter_u64_alloc(M_WAITOK);
1535 tcp_bad_csums = counter_u64_alloc(M_WAITOK);
1536 tcp_pacing_failures = counter_u64_alloc(M_WAITOK);
1537 tcp_dgp_failures = counter_u64_alloc(M_WAITOK);
1538
1539 hashsize = tcp_tcbhashsize;
1540 if (hashsize == 0) {
1541 /*
1542 * Auto tune the hash size based on maxsockets.
1543 * A perfect hash would have a 1:1 mapping
1544 * (hashsize = maxsockets) however it's been
1545 * suggested that O(2) average is better.
1546 */
1547 hashsize = maketcp_hashsize(maxsockets / 4);
1548 /*
1549 * Our historical default is 512,
1550 * do not autotune lower than this.
1551 */
1552 if (hashsize < 512)
1553 hashsize = 512;
1554 if (bootverbose)
1555 printf("%s: %s auto tuned to %d\n", __func__,
1556 "net.inet.tcp.tcbhashsize", hashsize);
1557 }
1558 /*
1559 * We require a hashsize to be a power of two.
1560 * Previously if it was not a power of two we would just reset it
1561 * back to 512, which could be a nasty surprise if you did not notice
1562 * the error message.
1563 * Instead what we do is clip it to the closest power of two lower
1564 * than the specified hash value.
1565 */
1566 if (!powerof2(hashsize)) {
1567 int oldhashsize = hashsize;
1568
1569 hashsize = maketcp_hashsize(hashsize);
1570 /* prevent absurdly low value */
1571 if (hashsize < 16)
1572 hashsize = 16;
1573 printf("%s: WARNING: TCB hash size not a power of 2, "
1574 "clipped from %d to %d.\n", __func__, oldhashsize,
1575 hashsize);
1576 }
1577 tcp_tcbhashsize = hashsize;
1578
1579 #ifdef INET
1580 IPPROTO_REGISTER(IPPROTO_TCP, tcp_input, tcp_ctlinput);
1581 #endif
1582 #ifdef INET6
1583 IP6PROTO_REGISTER(IPPROTO_TCP, tcp6_input, tcp6_ctlinput);
1584 #endif
1585 }
1586 SYSINIT(tcp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, tcp_init, NULL);
1587
1588 #ifdef VIMAGE
1589 static void
tcp_destroy(void * unused __unused)1590 tcp_destroy(void *unused __unused)
1591 {
1592 #ifdef TCP_HHOOK
1593 int error;
1594 #endif
1595
1596 tcp_hc_destroy();
1597 syncache_destroy();
1598 in_pcbinfo_destroy(&V_tcbinfo);
1599 /* tcp_discardcb() clears the sack_holes up. */
1600 uma_zdestroy(V_sack_hole_zone);
1601
1602 /*
1603 * Cannot free the zone until all tcpcbs are released as we attach
1604 * the allocations to them.
1605 */
1606 tcp_fastopen_destroy();
1607
1608 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES);
1609 VNET_PCPUSTAT_FREE(tcpstat);
1610
1611 #ifdef TCP_HHOOK
1612 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
1613 if (error != 0) {
1614 printf("%s: WARNING: unable to deregister helper hook "
1615 "type=%d, id=%d: error %d returned\n", __func__,
1616 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
1617 }
1618 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
1619 if (error != 0) {
1620 printf("%s: WARNING: unable to deregister helper hook "
1621 "type=%d, id=%d: error %d returned\n", __func__,
1622 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
1623 }
1624 #endif
1625 }
1626 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
1627 #endif
1628
1629 void
tcp_fini(void * xtp)1630 tcp_fini(void *xtp)
1631 {
1632
1633 }
1634
1635 /*
1636 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
1637 * tcp_template used to store this data in mbufs, but we now recopy it out
1638 * of the tcpcb each time to conserve mbufs.
1639 */
1640 void
tcpip_fillheaders(struct inpcb * inp,uint16_t port,void * ip_ptr,void * tcp_ptr)1641 tcpip_fillheaders(struct inpcb *inp, uint16_t port, void *ip_ptr, void *tcp_ptr)
1642 {
1643 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
1644
1645 INP_WLOCK_ASSERT(inp);
1646
1647 #ifdef INET6
1648 if ((inp->inp_vflag & INP_IPV6) != 0) {
1649 struct ip6_hdr *ip6;
1650
1651 ip6 = (struct ip6_hdr *)ip_ptr;
1652 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
1653 (inp->inp_flow & IPV6_FLOWINFO_MASK);
1654 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
1655 (IPV6_VERSION & IPV6_VERSION_MASK);
1656 if (port == 0)
1657 ip6->ip6_nxt = IPPROTO_TCP;
1658 else
1659 ip6->ip6_nxt = IPPROTO_UDP;
1660 ip6->ip6_plen = htons(sizeof(struct tcphdr));
1661 ip6->ip6_src = inp->in6p_laddr;
1662 ip6->ip6_dst = inp->in6p_faddr;
1663 }
1664 #endif /* INET6 */
1665 #if defined(INET6) && defined(INET)
1666 else
1667 #endif
1668 #ifdef INET
1669 {
1670 struct ip *ip;
1671
1672 ip = (struct ip *)ip_ptr;
1673 ip->ip_v = IPVERSION;
1674 ip->ip_hl = 5;
1675 ip->ip_tos = inp->inp_ip_tos;
1676 ip->ip_len = 0;
1677 ip->ip_id = 0;
1678 ip->ip_off = 0;
1679 ip->ip_ttl = inp->inp_ip_ttl;
1680 ip->ip_sum = 0;
1681 if (port == 0)
1682 ip->ip_p = IPPROTO_TCP;
1683 else
1684 ip->ip_p = IPPROTO_UDP;
1685 ip->ip_src = inp->inp_laddr;
1686 ip->ip_dst = inp->inp_faddr;
1687 }
1688 #endif /* INET */
1689 th->th_sport = inp->inp_lport;
1690 th->th_dport = inp->inp_fport;
1691 th->th_seq = 0;
1692 th->th_ack = 0;
1693 th->th_off = 5;
1694 tcp_set_flags(th, 0);
1695 th->th_win = 0;
1696 th->th_urp = 0;
1697 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
1698 }
1699
1700 /*
1701 * Create template to be used to send tcp packets on a connection.
1702 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
1703 * use for this function is in keepalives, which use tcp_respond.
1704 */
1705 struct tcptemp *
tcpip_maketemplate(struct inpcb * inp)1706 tcpip_maketemplate(struct inpcb *inp)
1707 {
1708 struct tcptemp *t;
1709
1710 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
1711 if (t == NULL)
1712 return (NULL);
1713 tcpip_fillheaders(inp, 0, (void *)&t->tt_ipgen, (void *)&t->tt_t);
1714 return (t);
1715 }
1716
1717 /*
1718 * Send a single message to the TCP at address specified by
1719 * the given TCP/IP header. If m == NULL, then we make a copy
1720 * of the tcpiphdr at th and send directly to the addressed host.
1721 * This is used to force keep alive messages out using the TCP
1722 * template for a connection. If flags are given then we send
1723 * a message back to the TCP which originated the segment th,
1724 * and discard the mbuf containing it and any other attached mbufs.
1725 *
1726 * In any case the ack and sequence number of the transmitted
1727 * segment are as specified by the parameters.
1728 *
1729 * NOTE: If m != NULL, then th must point to *inside* the mbuf.
1730 */
1731
1732 void
tcp_respond(struct tcpcb * tp,void * ipgen,struct tcphdr * th,struct mbuf * m,tcp_seq ack,tcp_seq seq,uint16_t flags)1733 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
1734 tcp_seq ack, tcp_seq seq, uint16_t flags)
1735 {
1736 struct tcpopt to;
1737 struct inpcb *inp;
1738 struct ip *ip;
1739 struct mbuf *optm;
1740 struct udphdr *uh = NULL;
1741 struct tcphdr *nth;
1742 struct tcp_log_buffer *lgb;
1743 u_char *optp;
1744 #ifdef INET6
1745 struct ip6_hdr *ip6;
1746 int isipv6;
1747 #endif /* INET6 */
1748 int optlen, tlen, win, ulen;
1749 int ect = 0;
1750 bool incl_opts;
1751 uint16_t port;
1752 int output_ret;
1753 #ifdef INVARIANTS
1754 int thflags = tcp_get_flags(th);
1755 #endif
1756
1757 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
1758 NET_EPOCH_ASSERT();
1759
1760 #ifdef INET6
1761 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
1762 ip6 = ipgen;
1763 #endif /* INET6 */
1764 ip = ipgen;
1765
1766 if (tp != NULL) {
1767 inp = tptoinpcb(tp);
1768 INP_LOCK_ASSERT(inp);
1769 } else
1770 inp = NULL;
1771
1772 if (m != NULL) {
1773 #ifdef INET6
1774 if (isipv6 && ip6 && (ip6->ip6_nxt == IPPROTO_UDP))
1775 port = m->m_pkthdr.tcp_tun_port;
1776 else
1777 #endif
1778 if (ip && (ip->ip_p == IPPROTO_UDP))
1779 port = m->m_pkthdr.tcp_tun_port;
1780 else
1781 port = 0;
1782 } else
1783 port = tp->t_port;
1784
1785 incl_opts = false;
1786 win = 0;
1787 if (tp != NULL) {
1788 if (!(flags & TH_RST)) {
1789 win = sbspace(&inp->inp_socket->so_rcv);
1790 if (win > TCP_MAXWIN << tp->rcv_scale)
1791 win = TCP_MAXWIN << tp->rcv_scale;
1792 }
1793 if ((tp->t_flags & TF_NOOPT) == 0)
1794 incl_opts = true;
1795 }
1796 if (m == NULL) {
1797 m = m_gethdr(M_NOWAIT, MT_DATA);
1798 if (m == NULL)
1799 return;
1800 m->m_data += max_linkhdr;
1801 #ifdef INET6
1802 if (isipv6) {
1803 bcopy((caddr_t)ip6, mtod(m, caddr_t),
1804 sizeof(struct ip6_hdr));
1805 ip6 = mtod(m, struct ip6_hdr *);
1806 nth = (struct tcphdr *)(ip6 + 1);
1807 if (port) {
1808 /* Insert a UDP header */
1809 uh = (struct udphdr *)nth;
1810 uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1811 uh->uh_dport = port;
1812 nth = (struct tcphdr *)(uh + 1);
1813 }
1814 } else
1815 #endif /* INET6 */
1816 {
1817 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1818 ip = mtod(m, struct ip *);
1819 nth = (struct tcphdr *)(ip + 1);
1820 if (port) {
1821 /* Insert a UDP header */
1822 uh = (struct udphdr *)nth;
1823 uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1824 uh->uh_dport = port;
1825 nth = (struct tcphdr *)(uh + 1);
1826 }
1827 }
1828 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1829 flags = TH_ACK;
1830 } else if ((!M_WRITABLE(m)) || (port != 0)) {
1831 struct mbuf *n;
1832
1833 /* Can't reuse 'm', allocate a new mbuf. */
1834 n = m_gethdr(M_NOWAIT, MT_DATA);
1835 if (n == NULL) {
1836 m_freem(m);
1837 return;
1838 }
1839
1840 if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1841 m_freem(m);
1842 m_freem(n);
1843 return;
1844 }
1845
1846 n->m_data += max_linkhdr;
1847 /* m_len is set later */
1848 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1849 #ifdef INET6
1850 if (isipv6) {
1851 bcopy((caddr_t)ip6, mtod(n, caddr_t),
1852 sizeof(struct ip6_hdr));
1853 ip6 = mtod(n, struct ip6_hdr *);
1854 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1855 nth = (struct tcphdr *)(ip6 + 1);
1856 if (port) {
1857 /* Insert a UDP header */
1858 uh = (struct udphdr *)nth;
1859 uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1860 uh->uh_dport = port;
1861 nth = (struct tcphdr *)(uh + 1);
1862 }
1863 } else
1864 #endif /* INET6 */
1865 {
1866 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1867 ip = mtod(n, struct ip *);
1868 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1869 nth = (struct tcphdr *)(ip + 1);
1870 if (port) {
1871 /* Insert a UDP header */
1872 uh = (struct udphdr *)nth;
1873 uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1874 uh->uh_dport = port;
1875 nth = (struct tcphdr *)(uh + 1);
1876 }
1877 }
1878 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1879 xchg(nth->th_dport, nth->th_sport, uint16_t);
1880 th = nth;
1881 m_freem(m);
1882 m = n;
1883 } else {
1884 /*
1885 * reuse the mbuf.
1886 * XXX MRT We inherit the FIB, which is lucky.
1887 */
1888 m_freem(m->m_next);
1889 m->m_next = NULL;
1890 m->m_data = (caddr_t)ipgen;
1891 /* clear any receive flags for proper bpf timestamping */
1892 m->m_flags &= ~(M_TSTMP | M_TSTMP_LRO);
1893 /* m_len is set later */
1894 #ifdef INET6
1895 if (isipv6) {
1896 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1897 nth = (struct tcphdr *)(ip6 + 1);
1898 } else
1899 #endif /* INET6 */
1900 {
1901 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1902 nth = (struct tcphdr *)(ip + 1);
1903 }
1904 if (th != nth) {
1905 /*
1906 * this is usually a case when an extension header
1907 * exists between the IPv6 header and the
1908 * TCP header.
1909 */
1910 nth->th_sport = th->th_sport;
1911 nth->th_dport = th->th_dport;
1912 }
1913 xchg(nth->th_dport, nth->th_sport, uint16_t);
1914 #undef xchg
1915 }
1916 tlen = 0;
1917 #ifdef INET6
1918 if (isipv6)
1919 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1920 #endif
1921 #if defined(INET) && defined(INET6)
1922 else
1923 #endif
1924 #ifdef INET
1925 tlen = sizeof (struct tcpiphdr);
1926 #endif
1927 if (port)
1928 tlen += sizeof (struct udphdr);
1929 #ifdef INVARIANTS
1930 m->m_len = 0;
1931 KASSERT(M_TRAILINGSPACE(m) >= tlen,
1932 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1933 m, tlen, (long)M_TRAILINGSPACE(m)));
1934 #endif
1935 m->m_len = tlen;
1936 to.to_flags = 0;
1937 if (incl_opts) {
1938 ect = tcp_ecn_output_established(tp, &flags, 0, false);
1939 /* Make sure we have room. */
1940 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1941 m->m_next = m_get(M_NOWAIT, MT_DATA);
1942 if (m->m_next) {
1943 optp = mtod(m->m_next, u_char *);
1944 optm = m->m_next;
1945 } else
1946 incl_opts = false;
1947 } else {
1948 optp = (u_char *) (nth + 1);
1949 optm = m;
1950 }
1951 }
1952 if (incl_opts) {
1953 /* Timestamps. */
1954 if (tp->t_flags & TF_RCVD_TSTMP) {
1955 to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1956 to.to_tsecr = tp->ts_recent;
1957 to.to_flags |= TOF_TS;
1958 }
1959 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1960 /* TCP-MD5 (RFC2385). */
1961 if (tp->t_flags & TF_SIGNATURE)
1962 to.to_flags |= TOF_SIGNATURE;
1963 #endif
1964 /* Add the options. */
1965 tlen += optlen = tcp_addoptions(&to, optp);
1966
1967 /* Update m_len in the correct mbuf. */
1968 optm->m_len += optlen;
1969 } else
1970 optlen = 0;
1971 #ifdef INET6
1972 if (isipv6) {
1973 if (uh) {
1974 ulen = tlen - sizeof(struct ip6_hdr);
1975 uh->uh_ulen = htons(ulen);
1976 }
1977 ip6->ip6_flow = htonl(ect << IPV6_FLOWLABEL_LEN);
1978 ip6->ip6_vfc = IPV6_VERSION;
1979 if (port)
1980 ip6->ip6_nxt = IPPROTO_UDP;
1981 else
1982 ip6->ip6_nxt = IPPROTO_TCP;
1983 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1984 }
1985 #endif
1986 #if defined(INET) && defined(INET6)
1987 else
1988 #endif
1989 #ifdef INET
1990 {
1991 if (uh) {
1992 ulen = tlen - sizeof(struct ip);
1993 uh->uh_ulen = htons(ulen);
1994 }
1995 ip->ip_len = htons(tlen);
1996 if (inp != NULL) {
1997 ip->ip_tos = inp->inp_ip_tos & ~IPTOS_ECN_MASK;
1998 ip->ip_ttl = inp->inp_ip_ttl;
1999 } else {
2000 ip->ip_tos = 0;
2001 ip->ip_ttl = V_ip_defttl;
2002 }
2003 ip->ip_tos |= ect;
2004 if (port) {
2005 ip->ip_p = IPPROTO_UDP;
2006 } else {
2007 ip->ip_p = IPPROTO_TCP;
2008 }
2009 if (V_path_mtu_discovery)
2010 ip->ip_off |= htons(IP_DF);
2011 }
2012 #endif
2013 m->m_pkthdr.len = tlen;
2014 m->m_pkthdr.rcvif = NULL;
2015 #ifdef MAC
2016 if (inp != NULL) {
2017 /*
2018 * Packet is associated with a socket, so allow the
2019 * label of the response to reflect the socket label.
2020 */
2021 INP_LOCK_ASSERT(inp);
2022 mac_inpcb_create_mbuf(inp, m);
2023 } else {
2024 /*
2025 * Packet is not associated with a socket, so possibly
2026 * update the label in place.
2027 */
2028 mac_netinet_tcp_reply(m);
2029 }
2030 #endif
2031 nth->th_seq = htonl(seq);
2032 nth->th_ack = htonl(ack);
2033 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
2034 tcp_set_flags(nth, flags);
2035 if (tp && (flags & TH_RST)) {
2036 /* Log the reset */
2037 tcp_log_end_status(tp, TCP_EI_STATUS_SERVER_RST);
2038 }
2039 if (tp != NULL)
2040 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
2041 else
2042 nth->th_win = htons((u_short)win);
2043 nth->th_urp = 0;
2044
2045 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2046 if (to.to_flags & TOF_SIGNATURE) {
2047 if (!TCPMD5_ENABLED() ||
2048 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
2049 m_freem(m);
2050 return;
2051 }
2052 }
2053 #endif
2054
2055 #ifdef INET6
2056 if (isipv6) {
2057 if (port) {
2058 m->m_pkthdr.csum_flags = CSUM_UDP_IPV6;
2059 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
2060 uh->uh_sum = in6_cksum_pseudo(ip6, ulen, IPPROTO_UDP, 0);
2061 nth->th_sum = 0;
2062 } else {
2063 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
2064 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
2065 nth->th_sum = in6_cksum_pseudo(ip6,
2066 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
2067 }
2068 ip6->ip6_hlim = in6_selecthlim(inp, NULL);
2069 }
2070 #endif /* INET6 */
2071 #if defined(INET6) && defined(INET)
2072 else
2073 #endif
2074 #ifdef INET
2075 {
2076 if (port) {
2077 uh->uh_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
2078 htons(ulen + IPPROTO_UDP));
2079 m->m_pkthdr.csum_flags = CSUM_UDP;
2080 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
2081 nth->th_sum = 0;
2082 } else {
2083 m->m_pkthdr.csum_flags = CSUM_TCP;
2084 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
2085 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
2086 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
2087 }
2088 }
2089 #endif /* INET */
2090 TCP_PROBE3(debug__output, tp, th, m);
2091 if (flags & TH_RST)
2092 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
2093 lgb = NULL;
2094 if ((tp != NULL) && tcp_bblogging_on(tp)) {
2095 if (INP_WLOCKED(inp)) {
2096 union tcp_log_stackspecific log;
2097 struct timeval tv;
2098
2099 memset(&log, 0, sizeof(log));
2100 log.u_bbr.inhpts = tcp_in_hpts(tp);
2101 log.u_bbr.flex8 = 4;
2102 log.u_bbr.pkts_out = tp->t_maxseg;
2103 log.u_bbr.timeStamp = tcp_get_usecs(&tv);
2104 log.u_bbr.delivered = 0;
2105 lgb = tcp_log_event(tp, nth, NULL, NULL, TCP_LOG_OUT,
2106 ERRNO_UNK, 0, &log, false, NULL, NULL, 0, &tv);
2107 } else {
2108 /*
2109 * We can not log the packet, since we only own the
2110 * read lock, but a write lock is needed. The read lock
2111 * is not upgraded to a write lock, since only getting
2112 * the read lock was done intentionally to improve the
2113 * handling of SYN flooding attacks.
2114 * This happens only for pure SYN segments received in
2115 * the initial CLOSED state, or received in a more
2116 * advanced state than listen and the UDP encapsulation
2117 * port is unexpected.
2118 * The incoming SYN segments do not really belong to
2119 * the TCP connection and the handling does not change
2120 * the state of the TCP connection. Therefore, the
2121 * sending of the RST segments is not logged. Please
2122 * note that also the incoming SYN segments are not
2123 * logged.
2124 *
2125 * The following code ensures that the above description
2126 * is and stays correct.
2127 */
2128 KASSERT((thflags & (TH_ACK|TH_SYN)) == TH_SYN &&
2129 (tp->t_state == TCPS_CLOSED ||
2130 (tp->t_state > TCPS_LISTEN && tp->t_port != port)),
2131 ("%s: Logging of TCP segment with flags 0x%b and "
2132 "UDP encapsulation port %u skipped in state %s",
2133 __func__, thflags, PRINT_TH_FLAGS,
2134 ntohs(port), tcpstates[tp->t_state]));
2135 }
2136 }
2137
2138 if (flags & TH_ACK)
2139 TCPSTAT_INC(tcps_sndacks);
2140 else if (flags & (TH_SYN|TH_FIN|TH_RST))
2141 TCPSTAT_INC(tcps_sndctrl);
2142 TCPSTAT_INC(tcps_sndtotal);
2143
2144 #ifdef INET6
2145 if (isipv6) {
2146 TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
2147 output_ret = ip6_output(m, inp ? inp->in6p_outputopts : NULL,
2148 NULL, 0, NULL, NULL, inp);
2149 }
2150 #endif /* INET6 */
2151 #if defined(INET) && defined(INET6)
2152 else
2153 #endif
2154 #ifdef INET
2155 {
2156 TCP_PROBE5(send, NULL, tp, ip, tp, nth);
2157 output_ret = ip_output(m, NULL, NULL, 0, NULL, inp);
2158 }
2159 #endif
2160 if (lgb != NULL)
2161 lgb->tlb_errno = output_ret;
2162 }
2163
2164 /*
2165 * Check that no more than V_tcp_ack_war_cnt per V_tcp_ack_war_time_window
2166 * are sent. *epoch_end is the end of the current epoch and is updated, if the
2167 * current epoch ended in the past. *ack_cnt is the counter used during the
2168 * current epoch. It might be reset and incremented.
2169 * The function returns true if a challenge ACK should be sent.
2170 */
2171 bool
tcp_challenge_ack_check(sbintime_t * epoch_end,uint32_t * ack_cnt)2172 tcp_challenge_ack_check(sbintime_t *epoch_end, uint32_t *ack_cnt)
2173 {
2174 sbintime_t now;
2175
2176 /*
2177 * The sending of a challenge ACK could be triggered by a blind attacker
2178 * to detect an existing TCP connection. To mitigate that, increment
2179 * also the global counter which would be incremented if the attacker
2180 * would have guessed wrongly.
2181 */
2182 (void)badport_bandlim(BANDLIM_TCP_RST);
2183
2184 if (V_tcp_ack_war_time_window == 0 || V_tcp_ack_war_cnt == 0) {
2185 /* ACK war protection is disabled. */
2186 return (true);
2187 } else {
2188 /* Start new epoch, if the previous one is already over. */
2189 now = getsbinuptime();
2190 if (*epoch_end < now) {
2191 *ack_cnt = 0;
2192 *epoch_end = now + V_tcp_ack_war_time_window * SBT_1MS;
2193 }
2194 /*
2195 * Send a challenge ACK, if less than tcp_ack_war_cnt have been
2196 * sent in the current epoch.
2197 */
2198 if (*ack_cnt < V_tcp_ack_war_cnt) {
2199 (*ack_cnt)++;
2200 return (true);
2201 } else {
2202 return (false);
2203 }
2204 }
2205 }
2206
2207 /*
2208 * Send a challenge ack (no data, no SACK option), but not more than
2209 * V_tcp_ack_war_cnt per V_tcp_ack_war_time_window (per TCP connection).
2210 */
2211 void
tcp_send_challenge_ack(struct tcpcb * tp,struct tcphdr * th,struct mbuf * m)2212 tcp_send_challenge_ack(struct tcpcb *tp, struct tcphdr *th, struct mbuf *m)
2213 {
2214 if (tcp_challenge_ack_check(&tp->t_challenge_ack_end,
2215 &tp->t_challenge_ack_cnt)) {
2216 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2217 tp->snd_nxt, TH_ACK);
2218 tp->last_ack_sent = tp->rcv_nxt;
2219 } else {
2220 m_freem(m);
2221 }
2222 }
2223
2224 /*
2225 * Create a new TCP control block, making an empty reassembly queue and hooking
2226 * it to the argument protocol control block. The `inp' parameter must have
2227 * come from the zone allocator set up by tcpcbstor declaration.
2228 * The caller can provide a pointer to a tcpcb of the listener to inherit the
2229 * TCP function block from the listener.
2230 */
2231 struct tcpcb *
tcp_newtcpcb(struct inpcb * inp,struct tcpcb * listening_tcb)2232 tcp_newtcpcb(struct inpcb *inp, struct tcpcb *listening_tcb)
2233 {
2234 struct tcpcb *tp = intotcpcb(inp);
2235 #ifdef INET6
2236 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
2237 #endif /* INET6 */
2238
2239 /*
2240 * Historically allocation was done with M_ZERO. There is a lot of
2241 * code that rely on that. For now take safe approach and zero whole
2242 * tcpcb. This definitely can be optimized.
2243 */
2244 bzero(&tp->t_start_zero, t_zero_size);
2245
2246 /* Initialise cc_var struct for this tcpcb. */
2247 tp->t_ccv.tp = tp;
2248 rw_rlock(&tcp_function_lock);
2249 if (listening_tcb != NULL) {
2250 INP_LOCK_ASSERT(tptoinpcb(listening_tcb));
2251 KASSERT(listening_tcb->t_fb != NULL,
2252 ("tcp_newtcpcb: listening_tcb->t_fb is NULL"));
2253 if (listening_tcb->t_fb->tfb_flags & TCP_FUNC_BEING_REMOVED) {
2254 rw_runlock(&tcp_function_lock);
2255 return (NULL);
2256 }
2257 tp->t_fb = listening_tcb->t_fb;
2258 } else {
2259 tp->t_fb = V_tcp_func_set_ptr;
2260 }
2261 refcount_acquire(&tp->t_fb->tfb_refcnt);
2262 KASSERT((tp->t_fb->tfb_flags & TCP_FUNC_BEING_REMOVED) == 0,
2263 ("tcp_newtcpcb: using TFB being removed"));
2264 rw_runlock(&tcp_function_lock);
2265 CC_LIST_RLOCK();
2266 if (listening_tcb != NULL) {
2267 if (CC_ALGO(listening_tcb)->flags & CC_MODULE_BEING_REMOVED) {
2268 CC_LIST_RUNLOCK();
2269 if (tp->t_fb->tfb_tcp_fb_fini)
2270 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2271 refcount_release(&tp->t_fb->tfb_refcnt);
2272 return (NULL);
2273 }
2274 CC_ALGO(tp) = CC_ALGO(listening_tcb);
2275 } else
2276 CC_ALGO(tp) = CC_DEFAULT_ALGO();
2277 cc_refer(CC_ALGO(tp));
2278 CC_LIST_RUNLOCK();
2279 if (CC_ALGO(tp)->cb_init != NULL)
2280 if (CC_ALGO(tp)->cb_init(&tp->t_ccv, NULL) > 0) {
2281 cc_detach(tp);
2282 if (tp->t_fb->tfb_tcp_fb_fini)
2283 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2284 refcount_release(&tp->t_fb->tfb_refcnt);
2285 return (NULL);
2286 }
2287
2288 #ifdef TCP_HHOOK
2289 if (khelp_init_osd(HELPER_CLASS_TCP, &tp->t_osd)) {
2290 if (CC_ALGO(tp)->cb_destroy != NULL)
2291 CC_ALGO(tp)->cb_destroy(&tp->t_ccv);
2292 CC_DATA(tp) = NULL;
2293 cc_detach(tp);
2294 if (tp->t_fb->tfb_tcp_fb_fini)
2295 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2296 refcount_release(&tp->t_fb->tfb_refcnt);
2297 return (NULL);
2298 }
2299 #endif
2300
2301 TAILQ_INIT(&tp->t_segq);
2302 STAILQ_INIT(&tp->t_inqueue);
2303 tp->t_maxseg =
2304 #ifdef INET6
2305 isipv6 ? V_tcp_v6mssdflt :
2306 #endif /* INET6 */
2307 V_tcp_mssdflt;
2308
2309 /* All mbuf queue/ack compress flags should be off */
2310 tcp_lro_features_off(tp);
2311
2312 tp->t_hpts_cpu = HPTS_CPU_NONE;
2313 tp->t_lro_cpu = HPTS_CPU_NONE;
2314
2315 callout_init_rw(&tp->t_callout, &inp->inp_lock,
2316 CALLOUT_TRYLOCK | CALLOUT_RETURNUNLOCKED);
2317 for (int i = 0; i < TT_N; i++)
2318 tp->t_timers[i] = SBT_MAX;
2319
2320 switch (V_tcp_do_rfc1323) {
2321 case 0:
2322 break;
2323 default:
2324 case 1:
2325 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
2326 break;
2327 case 2:
2328 tp->t_flags = TF_REQ_SCALE;
2329 break;
2330 case 3:
2331 tp->t_flags = TF_REQ_TSTMP;
2332 break;
2333 }
2334 if (V_tcp_do_sack)
2335 tp->t_flags |= TF_SACK_PERMIT;
2336 TAILQ_INIT(&tp->snd_holes);
2337
2338 /*
2339 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
2340 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
2341 * reasonable initial retransmit time.
2342 */
2343 tp->t_srtt = TCPTV_SRTTBASE;
2344 tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
2345 tp->t_rttmin = tcp_rexmit_min;
2346 tp->t_rxtcur = tcp_rexmit_initial;
2347 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2348 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2349 tp->t_rcvtime = ticks;
2350 /* We always start with ticks granularity */
2351 tp->t_tmr_granularity = TCP_TMR_GRANULARITY_TICKS;
2352 /*
2353 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
2354 * because the socket may be bound to an IPv6 wildcard address,
2355 * which may match an IPv4-mapped IPv6 address.
2356 */
2357 inp->inp_ip_ttl = V_ip_defttl;
2358 #ifdef TCP_BLACKBOX
2359 /* Initialize the per-TCPCB log data. */
2360 tcp_log_tcpcbinit(tp);
2361 #endif
2362 tp->t_pacing_rate = -1;
2363 if (tp->t_fb->tfb_tcp_fb_init) {
2364 if ((*tp->t_fb->tfb_tcp_fb_init)(tp, &tp->t_fb_ptr)) {
2365 if (CC_ALGO(tp)->cb_destroy != NULL)
2366 CC_ALGO(tp)->cb_destroy(&tp->t_ccv);
2367 CC_DATA(tp) = NULL;
2368 cc_detach(tp);
2369 #ifdef TCP_HHOOK
2370 khelp_destroy_osd(&tp->t_osd);
2371 #endif
2372 refcount_release(&tp->t_fb->tfb_refcnt);
2373 return (NULL);
2374 }
2375 }
2376 #ifdef STATS
2377 if (V_tcp_perconn_stats_enable == 1)
2378 tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0);
2379 #endif
2380 if (V_tcp_do_lrd)
2381 tp->t_flags |= TF_LRD;
2382
2383 return (tp);
2384 }
2385
2386 /*
2387 * Drop a TCP connection, reporting
2388 * the specified error. If connection is synchronized,
2389 * then send a RST to peer.
2390 */
2391 struct tcpcb *
tcp_drop(struct tcpcb * tp,int errno)2392 tcp_drop(struct tcpcb *tp, int errno)
2393 {
2394 struct socket *so = tptosocket(tp);
2395
2396 NET_EPOCH_ASSERT();
2397 INP_WLOCK_ASSERT(tptoinpcb(tp));
2398
2399 if (TCPS_HAVERCVDSYN(tp->t_state)) {
2400 tcp_state_change(tp, TCPS_CLOSED);
2401 /* Don't use tcp_output() here due to possible recursion. */
2402 (void)tcp_output_nodrop(tp);
2403 TCPSTAT_INC(tcps_drops);
2404 } else
2405 TCPSTAT_INC(tcps_conndrops);
2406 if (errno == ETIMEDOUT && tp->t_softerror)
2407 errno = tp->t_softerror;
2408 so->so_error = errno;
2409 return (tcp_close(tp));
2410 }
2411
2412 void
tcp_discardcb(struct tcpcb * tp)2413 tcp_discardcb(struct tcpcb *tp)
2414 {
2415 struct inpcb *inp = tptoinpcb(tp);
2416 struct socket *so = tptosocket(tp);
2417 struct mbuf *m;
2418 #ifdef INET6
2419 bool isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
2420 #endif
2421
2422 INP_WLOCK_ASSERT(inp);
2423 MPASS(!callout_active(&tp->t_callout));
2424 MPASS(TAILQ_EMPTY(&tp->snd_holes));
2425
2426 /* free the reassembly queue, if any */
2427 tcp_reass_flush(tp);
2428
2429 #ifdef TCP_OFFLOAD
2430 /* Disconnect offload device, if any. */
2431 if (tp->t_flags & TF_TOE)
2432 tcp_offload_detach(tp);
2433 #endif
2434
2435 /* Allow the CC algorithm to clean up after itself. */
2436 if (CC_ALGO(tp)->cb_destroy != NULL)
2437 CC_ALGO(tp)->cb_destroy(&tp->t_ccv);
2438 CC_DATA(tp) = NULL;
2439 /* Detach from the CC algorithm */
2440 cc_detach(tp);
2441
2442 #ifdef TCP_HHOOK
2443 khelp_destroy_osd(&tp->t_osd);
2444 #endif
2445 #ifdef STATS
2446 stats_blob_destroy(tp->t_stats);
2447 #endif
2448
2449 CC_ALGO(tp) = NULL;
2450 if ((m = STAILQ_FIRST(&tp->t_inqueue)) != NULL) {
2451 struct mbuf *prev;
2452
2453 STAILQ_INIT(&tp->t_inqueue);
2454 STAILQ_FOREACH_FROM_SAFE(m, &tp->t_inqueue, m_stailqpkt, prev)
2455 m_freem(m);
2456 }
2457 TCPSTATES_DEC(tp->t_state);
2458
2459 if (tp->t_fb->tfb_tcp_fb_fini)
2460 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2461 MPASS(!tcp_in_hpts(tp));
2462 #ifdef TCP_BLACKBOX
2463 tcp_log_tcpcbfini(tp);
2464 #endif
2465
2466 /*
2467 * If we got enough samples through the srtt filter,
2468 * save the rtt and rttvar in the routing entry.
2469 * 'Enough' is arbitrarily defined as 4 rtt samples.
2470 * 4 samples is enough for the srtt filter to converge
2471 * to within enough % of the correct value; fewer samples
2472 * and we could save a bogus rtt. The danger is not high
2473 * as tcp quickly recovers from everything.
2474 * XXX: Works very well but needs some more statistics!
2475 *
2476 * XXXRRS: Updating must be after the stack fini() since
2477 * that may be converting some internal representation of
2478 * say srtt etc into the general one used by other stacks.
2479 */
2480 if (tp->t_rttupdated >= 4) {
2481 struct hc_metrics_lite metrics;
2482 uint32_t ssthresh;
2483
2484 bzero(&metrics, sizeof(metrics));
2485 /*
2486 * Update the ssthresh always when the conditions below
2487 * are satisfied. This gives us better new start value
2488 * for the congestion avoidance for new connections.
2489 * ssthresh is only set if packet loss occurred on a session.
2490 */
2491 ssthresh = tp->snd_ssthresh;
2492 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
2493 /*
2494 * convert the limit from user data bytes to
2495 * packets then to packet data bytes.
2496 */
2497 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
2498 if (ssthresh < 2)
2499 ssthresh = 2;
2500 ssthresh *= (tp->t_maxseg +
2501 #ifdef INET6
2502 (isipv6 ? sizeof (struct ip6_hdr) +
2503 sizeof (struct tcphdr) :
2504 #endif
2505 sizeof (struct tcpiphdr)
2506 #ifdef INET6
2507 )
2508 #endif
2509 );
2510 } else
2511 ssthresh = 0;
2512 metrics.hc_ssthresh = ssthresh;
2513
2514 metrics.hc_rtt = tp->t_srtt;
2515 metrics.hc_rttvar = tp->t_rttvar;
2516 metrics.hc_cwnd = tp->snd_cwnd;
2517 metrics.hc_sendpipe = 0;
2518 metrics.hc_recvpipe = 0;
2519
2520 tcp_hc_update(&inp->inp_inc, &metrics);
2521 }
2522
2523 refcount_release(&tp->t_fb->tfb_refcnt);
2524 }
2525
2526 /*
2527 * Attempt to close a TCP control block, marking it as dropped, and freeing
2528 * the socket if we hold the only reference.
2529 */
2530 struct tcpcb *
tcp_close(struct tcpcb * tp)2531 tcp_close(struct tcpcb *tp)
2532 {
2533 struct inpcb *inp = tptoinpcb(tp);
2534 struct socket *so = tptosocket(tp);
2535
2536 INP_WLOCK_ASSERT(inp);
2537
2538 #ifdef TCP_OFFLOAD
2539 if (tp->t_state == TCPS_LISTEN)
2540 tcp_offload_listen_stop(tp);
2541 #endif
2542 /*
2543 * This releases the TFO pending counter resource for TFO listen
2544 * sockets as well as passively-created TFO sockets that transition
2545 * from SYN_RECEIVED to CLOSED.
2546 */
2547 if (tp->t_tfo_pending) {
2548 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2549 tp->t_tfo_pending = NULL;
2550 }
2551 tcp_timer_stop(tp);
2552 if (tp->t_fb->tfb_tcp_timer_stop_all != NULL)
2553 tp->t_fb->tfb_tcp_timer_stop_all(tp);
2554 in_pcbdrop(inp);
2555 TCPSTAT_INC(tcps_closed);
2556 if (tp->t_state != TCPS_CLOSED)
2557 tcp_state_change(tp, TCPS_CLOSED);
2558 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
2559 tcp_free_sackholes(tp);
2560 soisdisconnected(so);
2561 if (inp->inp_flags & INP_SOCKREF) {
2562 inp->inp_flags &= ~INP_SOCKREF;
2563 INP_WUNLOCK(inp);
2564 sorele(so);
2565 return (NULL);
2566 }
2567 return (tp);
2568 }
2569
2570 /*
2571 * Notify a tcp user of an asynchronous error;
2572 * store error as soft error, but wake up user
2573 * (for now, won't do anything until can select for soft error).
2574 *
2575 * Do not wake up user since there currently is no mechanism for
2576 * reporting soft errors (yet - a kqueue filter may be added).
2577 */
2578 static struct inpcb *
tcp_notify(struct inpcb * inp,int error)2579 tcp_notify(struct inpcb *inp, int error)
2580 {
2581 struct tcpcb *tp;
2582
2583 INP_WLOCK_ASSERT(inp);
2584
2585 tp = intotcpcb(inp);
2586 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
2587
2588 /*
2589 * Ignore some errors if we are hooked up.
2590 * If connection hasn't completed, has retransmitted several times,
2591 * and receives a second error, give up now. This is better
2592 * than waiting a long time to establish a connection that
2593 * can never complete.
2594 */
2595 if (tp->t_state == TCPS_ESTABLISHED &&
2596 (error == EHOSTUNREACH || error == ENETUNREACH ||
2597 error == EHOSTDOWN)) {
2598 if (inp->inp_route.ro_nh) {
2599 NH_FREE(inp->inp_route.ro_nh);
2600 inp->inp_route.ro_nh = (struct nhop_object *)NULL;
2601 }
2602 return (inp);
2603 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
2604 tp->t_softerror) {
2605 tp = tcp_drop(tp, error);
2606 if (tp != NULL)
2607 return (inp);
2608 else
2609 return (NULL);
2610 } else {
2611 tp->t_softerror = error;
2612 return (inp);
2613 }
2614 #if 0
2615 wakeup( &so->so_timeo);
2616 sorwakeup(so);
2617 sowwakeup(so);
2618 #endif
2619 }
2620
2621 static int
tcp_pcblist(SYSCTL_HANDLER_ARGS)2622 tcp_pcblist(SYSCTL_HANDLER_ARGS)
2623 {
2624 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
2625 INPLOOKUP_RLOCKPCB);
2626 struct xinpgen xig;
2627 struct inpcb *inp;
2628 int error;
2629
2630 if (req->newptr != NULL)
2631 return (EPERM);
2632
2633 if (req->oldptr == NULL) {
2634 int n;
2635
2636 n = V_tcbinfo.ipi_count +
2637 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2638 n += imax(n / 8, 10);
2639 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
2640 return (0);
2641 }
2642
2643 if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
2644 return (error);
2645
2646 bzero(&xig, sizeof(xig));
2647 xig.xig_len = sizeof xig;
2648 xig.xig_count = V_tcbinfo.ipi_count +
2649 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2650 xig.xig_gen = V_tcbinfo.ipi_gencnt;
2651 xig.xig_sogen = so_gencnt;
2652 error = SYSCTL_OUT(req, &xig, sizeof xig);
2653 if (error)
2654 return (error);
2655
2656 error = syncache_pcblist(req);
2657 if (error)
2658 return (error);
2659
2660 while ((inp = inp_next(&inpi)) != NULL) {
2661 if (inp->inp_gencnt <= xig.xig_gen &&
2662 cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
2663 struct xtcpcb xt;
2664
2665 tcp_inptoxtp(inp, &xt);
2666 error = SYSCTL_OUT(req, &xt, sizeof xt);
2667 if (error) {
2668 INP_RUNLOCK(inp);
2669 break;
2670 } else
2671 continue;
2672 }
2673 }
2674
2675 if (!error) {
2676 /*
2677 * Give the user an updated idea of our state.
2678 * If the generation differs from what we told
2679 * her before, she knows that something happened
2680 * while we were processing this request, and it
2681 * might be necessary to retry.
2682 */
2683 xig.xig_gen = V_tcbinfo.ipi_gencnt;
2684 xig.xig_sogen = so_gencnt;
2685 xig.xig_count = V_tcbinfo.ipi_count +
2686 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2687 error = SYSCTL_OUT(req, &xig, sizeof xig);
2688 }
2689
2690 return (error);
2691 }
2692
2693 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
2694 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
2695 NULL, 0, tcp_pcblist, "S,xtcpcb",
2696 "List of active TCP connections");
2697
2698 #define SND_TAG_STATUS_MAXLEN 128
2699
2700 #ifdef KERN_TLS
2701
2702 static struct sx ktlslist_lock;
2703 SX_SYSINIT(ktlslistlock, &ktlslist_lock, "ktlslist");
2704 static uint64_t ktls_glob_gen = 1;
2705
2706 static int
tcp_ktlslist_locked(SYSCTL_HANDLER_ARGS,bool export_keys)2707 tcp_ktlslist_locked(SYSCTL_HANDLER_ARGS, bool export_keys)
2708 {
2709 struct xinpgen xig;
2710 struct inpcb *inp;
2711 struct socket *so;
2712 struct ktls_session *ksr, *kss;
2713 char *buf;
2714 struct xktls_session *xktls;
2715 uint64_t ipi_gencnt;
2716 size_t buflen, len, sz;
2717 u_int cnt;
2718 int error;
2719 bool ek, p;
2720
2721 sx_assert(&ktlslist_lock, SA_XLOCKED);
2722 if (req->newptr != NULL)
2723 return (EPERM);
2724
2725 len = 0;
2726 cnt = 0;
2727 ipi_gencnt = V_tcbinfo.ipi_gencnt;
2728 bzero(&xig, sizeof(xig));
2729 xig.xig_len = sizeof(xig);
2730 xig.xig_gen = ktls_glob_gen++;
2731 xig.xig_sogen = so_gencnt;
2732
2733 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
2734 INPLOOKUP_RLOCKPCB);
2735 while ((inp = inp_next(&inpi)) != NULL) {
2736 if (inp->inp_gencnt > ipi_gencnt ||
2737 cr_canseeinpcb(req->td->td_ucred, inp) != 0)
2738 continue;
2739
2740 so = inp->inp_socket;
2741 if (so != NULL && so->so_gencnt <= xig.xig_sogen) {
2742 p = false;
2743 ek = export_keys && cr_canexport_ktlskeys(
2744 req->td, inp);
2745 ksr = so->so_rcv.sb_tls_info;
2746 if (ksr != NULL) {
2747 ksr->gen = xig.xig_gen;
2748 p = true;
2749 if (ek) {
2750 sz = SIZE_T_MAX;
2751 ktls_session_copy_keys(ksr,
2752 NULL, &sz);
2753 len += sz;
2754 }
2755 if (ksr->snd_tag != NULL &&
2756 ksr->snd_tag->sw->snd_tag_status_str !=
2757 NULL) {
2758 sz = SND_TAG_STATUS_MAXLEN;
2759 in_pcbref(inp);
2760 INP_RUNLOCK(inp);
2761 error = ksr->snd_tag->sw->
2762 snd_tag_status_str(
2763 ksr->snd_tag, NULL, &sz);
2764 if (in_pcbrele_rlock(inp))
2765 return (EDEADLK);
2766 if (error == 0)
2767 len += sz;
2768 }
2769 }
2770 kss = so->so_snd.sb_tls_info;
2771 if (kss != NULL) {
2772 kss->gen = xig.xig_gen;
2773 p = true;
2774 if (ek) {
2775 sz = SIZE_T_MAX;
2776 ktls_session_copy_keys(kss,
2777 NULL, &sz);
2778 len += sz;
2779 }
2780 if (kss->snd_tag != NULL &&
2781 kss->snd_tag->sw->snd_tag_status_str !=
2782 NULL) {
2783 sz = SND_TAG_STATUS_MAXLEN;
2784 in_pcbref(inp);
2785 INP_RUNLOCK(inp);
2786 error = kss->snd_tag->sw->
2787 snd_tag_status_str(
2788 kss->snd_tag, NULL, &sz);
2789 if (in_pcbrele_rlock(inp))
2790 return (EDEADLK);
2791 if (error == 0)
2792 len += sz;
2793 }
2794 }
2795 if (p) {
2796 len += sizeof(*xktls);
2797 len = roundup2(len, __alignof(struct
2798 xktls_session));
2799 }
2800 }
2801 }
2802 if (req->oldptr == NULL) {
2803 len += 2 * sizeof(xig);
2804 len += 3 * len / 4;
2805 req->oldidx = len;
2806 return (0);
2807 }
2808
2809 if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
2810 return (error);
2811
2812 error = SYSCTL_OUT(req, &xig, sizeof xig);
2813 if (error != 0)
2814 return (error);
2815
2816 buflen = roundup2(sizeof(*xktls) + 2 * TLS_MAX_PARAM_SIZE +
2817 2 * SND_TAG_STATUS_MAXLEN, __alignof(struct xktls_session));
2818 buf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
2819 struct inpcb_iterator inpi1 = INP_ALL_ITERATOR(&V_tcbinfo,
2820 INPLOOKUP_RLOCKPCB);
2821 while ((inp = inp_next(&inpi1)) != NULL) {
2822 if (inp->inp_gencnt > ipi_gencnt ||
2823 cr_canseeinpcb(req->td->td_ucred, inp) != 0)
2824 continue;
2825
2826 so = inp->inp_socket;
2827 if (so == NULL)
2828 continue;
2829
2830 p = false;
2831 ek = export_keys && cr_canexport_ktlskeys(req->td, inp);
2832 ksr = so->so_rcv.sb_tls_info;
2833 kss = so->so_snd.sb_tls_info;
2834 xktls = (struct xktls_session *)buf;
2835 if (ksr != NULL && ksr->gen == xig.xig_gen) {
2836 p = true;
2837 ktls_session_to_xktls_onedir(ksr, ek, &xktls->rcv);
2838 }
2839 if (kss != NULL && kss->gen == xig.xig_gen) {
2840 p = true;
2841 ktls_session_to_xktls_onedir(kss, ek, &xktls->snd);
2842 }
2843 if (!p)
2844 continue;
2845
2846 xktls->inp_gencnt = inp->inp_gencnt;
2847 xktls->so_pcb = (kvaddr_t)inp;
2848 memcpy(&xktls->coninf, &inp->inp_inc, sizeof(xktls->coninf));
2849 len = sizeof(*xktls);
2850 if (ksr != NULL && ksr->gen == xig.xig_gen) {
2851 if (ek) {
2852 sz = buflen - len;
2853 ktls_session_copy_keys(ksr, buf + len, &sz);
2854 len += sz;
2855 } else {
2856 xktls->rcv.cipher_key_len = 0;
2857 xktls->rcv.auth_key_len = 0;
2858 }
2859 if (ksr->snd_tag != NULL &&
2860 ksr->snd_tag->sw->snd_tag_status_str != NULL) {
2861 sz = SND_TAG_STATUS_MAXLEN;
2862 in_pcbref(inp);
2863 INP_RUNLOCK(inp);
2864 error = ksr->snd_tag->sw->snd_tag_status_str(
2865 ksr->snd_tag, buf + len, &sz);
2866 if (in_pcbrele_rlock(inp))
2867 return (EDEADLK);
2868 if (error == 0) {
2869 xktls->rcv.drv_st_len = sz;
2870 len += sz;
2871 }
2872 }
2873 }
2874 if (kss != NULL && kss->gen == xig.xig_gen) {
2875 if (ek) {
2876 sz = buflen - len;
2877 ktls_session_copy_keys(kss, buf + len, &sz);
2878 len += sz;
2879 } else {
2880 xktls->snd.cipher_key_len = 0;
2881 xktls->snd.auth_key_len = 0;
2882 }
2883 if (kss->snd_tag != NULL &&
2884 kss->snd_tag->sw->snd_tag_status_str != NULL) {
2885 sz = SND_TAG_STATUS_MAXLEN;
2886 in_pcbref(inp);
2887 INP_RUNLOCK(inp);
2888 error = kss->snd_tag->sw->snd_tag_status_str(
2889 kss->snd_tag, buf + len, &sz);
2890 if (in_pcbrele_rlock(inp))
2891 return (EDEADLK);
2892 if (error == 0) {
2893 xktls->snd.drv_st_len = sz;
2894 len += sz;
2895 }
2896 }
2897 }
2898 len = roundup2(len, __alignof(*xktls));
2899 xktls->tsz = len;
2900 xktls->fsz = sizeof(*xktls);
2901
2902 error = SYSCTL_OUT(req, xktls, len);
2903 if (error != 0) {
2904 INP_RUNLOCK(inp);
2905 break;
2906 }
2907 cnt++;
2908 }
2909
2910 if (error == 0) {
2911 xig.xig_sogen = so_gencnt;
2912 xig.xig_count = cnt;
2913 error = SYSCTL_OUT(req, &xig, sizeof(xig));
2914 }
2915
2916 zfree(buf, M_TEMP);
2917 return (error);
2918 }
2919
2920 static int
tcp_ktlslist1(SYSCTL_HANDLER_ARGS,bool export_keys)2921 tcp_ktlslist1(SYSCTL_HANDLER_ARGS, bool export_keys)
2922 {
2923 int repeats, error;
2924
2925 for (repeats = 0; repeats < 100; repeats++) {
2926 if (sx_xlock_sig(&ktlslist_lock))
2927 return (EINTR);
2928 error = tcp_ktlslist_locked(oidp, arg1, arg2, req,
2929 export_keys);
2930 sx_xunlock(&ktlslist_lock);
2931 if (error != EDEADLK)
2932 break;
2933 if (sig_intr() != 0) {
2934 error = EINTR;
2935 break;
2936 }
2937 req->oldidx = 0;
2938 }
2939 return (error);
2940 }
2941
2942 static int
tcp_ktlslist_nokeys(SYSCTL_HANDLER_ARGS)2943 tcp_ktlslist_nokeys(SYSCTL_HANDLER_ARGS)
2944 {
2945 return (tcp_ktlslist1(oidp, arg1, arg2, req, false));
2946 }
2947
2948 static int
tcp_ktlslist_wkeys(SYSCTL_HANDLER_ARGS)2949 tcp_ktlslist_wkeys(SYSCTL_HANDLER_ARGS)
2950 {
2951 return (tcp_ktlslist1(oidp, arg1, arg2, req, true));
2952 }
2953
2954 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KTLSLIST, ktlslist,
2955 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2956 NULL, 0, tcp_ktlslist_nokeys, "S,xktls_session",
2957 "List of active kTLS sessions for TCP connections");
2958 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KTLSLIST_WKEYS, ktlslist_wkeys,
2959 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2960 NULL, 0, tcp_ktlslist_wkeys, "S,xktls_session",
2961 "List of active kTLS sessions for TCP connections with keys");
2962 #endif /* KERN_TLS */
2963
2964 #ifdef INET
2965 static int
tcp_getcred(SYSCTL_HANDLER_ARGS)2966 tcp_getcred(SYSCTL_HANDLER_ARGS)
2967 {
2968 struct xucred xuc;
2969 struct sockaddr_in addrs[2];
2970 struct epoch_tracker et;
2971 struct inpcb *inp;
2972 int error;
2973
2974 if (req->newptr == NULL)
2975 return (EINVAL);
2976 error = priv_check(req->td, PRIV_NETINET_GETCRED);
2977 if (error)
2978 return (error);
2979 error = SYSCTL_IN(req, addrs, sizeof(addrs));
2980 if (error)
2981 return (error);
2982 NET_EPOCH_ENTER(et);
2983 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
2984 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
2985 NET_EPOCH_EXIT(et);
2986 if (inp != NULL) {
2987 if (error == 0)
2988 error = cr_canseeinpcb(req->td->td_ucred, inp);
2989 if (error == 0)
2990 cru2x(inp->inp_cred, &xuc);
2991 INP_RUNLOCK(inp);
2992 } else
2993 error = ENOENT;
2994 if (error == 0)
2995 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2996 return (error);
2997 }
2998
2999 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
3000 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
3001 0, 0, tcp_getcred, "S,xucred",
3002 "Get the xucred of a TCP connection");
3003 #endif /* INET */
3004
3005 #ifdef INET6
3006 static int
tcp6_getcred(SYSCTL_HANDLER_ARGS)3007 tcp6_getcred(SYSCTL_HANDLER_ARGS)
3008 {
3009 struct epoch_tracker et;
3010 struct xucred xuc;
3011 struct sockaddr_in6 addrs[2];
3012 struct inpcb *inp;
3013 int error;
3014 #ifdef INET
3015 int mapped = 0;
3016 #endif
3017
3018 if (req->newptr == NULL)
3019 return (EINVAL);
3020 error = priv_check(req->td, PRIV_NETINET_GETCRED);
3021 if (error)
3022 return (error);
3023 error = SYSCTL_IN(req, addrs, sizeof(addrs));
3024 if (error)
3025 return (error);
3026 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
3027 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
3028 return (error);
3029 }
3030 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
3031 #ifdef INET
3032 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
3033 mapped = 1;
3034 else
3035 #endif
3036 return (EINVAL);
3037 }
3038
3039 NET_EPOCH_ENTER(et);
3040 #ifdef INET
3041 if (mapped == 1)
3042 inp = in_pcblookup(&V_tcbinfo,
3043 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
3044 addrs[1].sin6_port,
3045 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
3046 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
3047 else
3048 #endif
3049 inp = in6_pcblookup(&V_tcbinfo,
3050 &addrs[1].sin6_addr, addrs[1].sin6_port,
3051 &addrs[0].sin6_addr, addrs[0].sin6_port,
3052 INPLOOKUP_RLOCKPCB, NULL);
3053 NET_EPOCH_EXIT(et);
3054 if (inp != NULL) {
3055 if (error == 0)
3056 error = cr_canseeinpcb(req->td->td_ucred, inp);
3057 if (error == 0)
3058 cru2x(inp->inp_cred, &xuc);
3059 INP_RUNLOCK(inp);
3060 } else
3061 error = ENOENT;
3062 if (error == 0)
3063 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
3064 return (error);
3065 }
3066
3067 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
3068 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
3069 0, 0, tcp6_getcred, "S,xucred",
3070 "Get the xucred of a TCP6 connection");
3071 #endif /* INET6 */
3072
3073 #ifdef INET
3074 /* Path MTU to try next when a fragmentation-needed message is received. */
3075 static inline int
tcp_next_pmtu(const struct icmp * icp,const struct ip * ip)3076 tcp_next_pmtu(const struct icmp *icp, const struct ip *ip)
3077 {
3078 int mtu = ntohs(icp->icmp_nextmtu);
3079
3080 /* If no alternative MTU was proposed, try the next smaller one. */
3081 if (!mtu)
3082 mtu = ip_next_mtu(ntohs(ip->ip_len), 1);
3083 if (mtu < V_tcp_minmss + sizeof(struct tcpiphdr))
3084 mtu = V_tcp_minmss + sizeof(struct tcpiphdr);
3085
3086 return (mtu);
3087 }
3088
3089 static void
tcp_ctlinput_with_port(struct icmp * icp,uint16_t port)3090 tcp_ctlinput_with_port(struct icmp *icp, uint16_t port)
3091 {
3092 struct ip *ip;
3093 struct tcphdr *th;
3094 struct inpcb *inp;
3095 struct tcpcb *tp;
3096 struct inpcb *(*notify)(struct inpcb *, int);
3097 struct in_conninfo inc;
3098 tcp_seq icmp_tcp_seq;
3099 int errno, mtu;
3100
3101 errno = icmp_errmap(icp);
3102 switch (errno) {
3103 case 0:
3104 return;
3105 case EMSGSIZE:
3106 notify = tcp_mtudisc_notify;
3107 break;
3108 case ECONNREFUSED:
3109 if (V_icmp_may_rst)
3110 notify = tcp_drop_syn_sent;
3111 else
3112 notify = tcp_notify;
3113 break;
3114 case EHOSTUNREACH:
3115 if (V_icmp_may_rst && icp->icmp_type == ICMP_TIMXCEED)
3116 notify = tcp_drop_syn_sent;
3117 else
3118 notify = tcp_notify;
3119 break;
3120 default:
3121 notify = tcp_notify;
3122 }
3123
3124 ip = &icp->icmp_ip;
3125 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
3126 icmp_tcp_seq = th->th_seq;
3127 inp = in_pcblookup(&V_tcbinfo, ip->ip_dst, th->th_dport, ip->ip_src,
3128 th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
3129 if (inp != NULL) {
3130 tp = intotcpcb(inp);
3131 #ifdef TCP_OFFLOAD
3132 if (tp->t_flags & TF_TOE && errno == EMSGSIZE) {
3133 /*
3134 * MTU discovery for offloaded connections. Let
3135 * the TOE driver verify seq# and process it.
3136 */
3137 mtu = tcp_next_pmtu(icp, ip);
3138 tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu);
3139 goto out;
3140 }
3141 #endif
3142 if (tp->t_port != port)
3143 goto out;
3144 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
3145 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
3146 if (errno == EMSGSIZE) {
3147 /*
3148 * MTU discovery: we got a needfrag and
3149 * will potentially try a lower MTU.
3150 */
3151 mtu = tcp_next_pmtu(icp, ip);
3152
3153 /*
3154 * Only process the offered MTU if it
3155 * is smaller than the current one.
3156 */
3157 if (mtu < tp->t_maxseg +
3158 sizeof(struct tcpiphdr)) {
3159 bzero(&inc, sizeof(inc));
3160 inc.inc_faddr = ip->ip_dst;
3161 inc.inc_fibnum =
3162 inp->inp_inc.inc_fibnum;
3163 tcp_hc_updatemtu(&inc, mtu);
3164 inp = tcp_mtudisc(inp, mtu);
3165 }
3166 } else
3167 inp = (*notify)(inp, errno);
3168 }
3169 } else {
3170 bzero(&inc, sizeof(inc));
3171 inc.inc_fport = th->th_dport;
3172 inc.inc_lport = th->th_sport;
3173 inc.inc_faddr = ip->ip_dst;
3174 inc.inc_laddr = ip->ip_src;
3175 syncache_unreach(&inc, icmp_tcp_seq, port);
3176 }
3177 out:
3178 if (inp != NULL)
3179 INP_WUNLOCK(inp);
3180 }
3181
3182 static void
tcp_ctlinput(struct icmp * icmp)3183 tcp_ctlinput(struct icmp *icmp)
3184 {
3185 tcp_ctlinput_with_port(icmp, htons(0));
3186 }
3187
3188 static void
tcp_ctlinput_viaudp(udp_tun_icmp_param_t param)3189 tcp_ctlinput_viaudp(udp_tun_icmp_param_t param)
3190 {
3191 /* Its a tunneled TCP over UDP icmp */
3192 struct icmp *icmp = param.icmp;
3193 struct ip *outer_ip, *inner_ip;
3194 struct udphdr *udp;
3195 struct tcphdr *th, ttemp;
3196 int i_hlen, o_len;
3197 uint16_t port;
3198
3199 outer_ip = (struct ip *)((caddr_t)icmp - sizeof(struct ip));
3200 inner_ip = &icmp->icmp_ip;
3201 i_hlen = inner_ip->ip_hl << 2;
3202 o_len = ntohs(outer_ip->ip_len);
3203 if (o_len <
3204 (sizeof(struct ip) + 8 + i_hlen + sizeof(struct udphdr) + offsetof(struct tcphdr, th_ack))) {
3205 /* Not enough data present */
3206 return;
3207 }
3208 /* Ok lets strip out the inner udphdr header by copying up on top of it the tcp hdr */
3209 udp = (struct udphdr *)(((caddr_t)inner_ip) + i_hlen);
3210 if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) {
3211 return;
3212 }
3213 port = udp->uh_dport;
3214 th = (struct tcphdr *)(udp + 1);
3215 memcpy(&ttemp, th, sizeof(struct tcphdr));
3216 memcpy(udp, &ttemp, sizeof(struct tcphdr));
3217 /* Now adjust down the size of the outer IP header */
3218 o_len -= sizeof(struct udphdr);
3219 outer_ip->ip_len = htons(o_len);
3220 /* Now call in to the normal handling code */
3221 tcp_ctlinput_with_port(icmp, port);
3222 }
3223 #endif /* INET */
3224
3225 #ifdef INET6
3226 static inline int
tcp6_next_pmtu(const struct icmp6_hdr * icmp6)3227 tcp6_next_pmtu(const struct icmp6_hdr *icmp6)
3228 {
3229 int mtu = ntohl(icmp6->icmp6_mtu);
3230
3231 /*
3232 * If no alternative MTU was proposed, or the proposed MTU was too
3233 * small, set to the min.
3234 */
3235 if (mtu < IPV6_MMTU)
3236 mtu = IPV6_MMTU;
3237 return (mtu);
3238 }
3239
3240 static void
tcp6_ctlinput_with_port(struct ip6ctlparam * ip6cp,uint16_t port)3241 tcp6_ctlinput_with_port(struct ip6ctlparam *ip6cp, uint16_t port)
3242 {
3243 struct in6_addr *dst;
3244 struct inpcb *(*notify)(struct inpcb *, int);
3245 struct ip6_hdr *ip6;
3246 struct mbuf *m;
3247 struct inpcb *inp;
3248 struct tcpcb *tp;
3249 struct icmp6_hdr *icmp6;
3250 struct in_conninfo inc;
3251 struct tcp_ports {
3252 uint16_t th_sport;
3253 uint16_t th_dport;
3254 } t_ports;
3255 tcp_seq icmp_tcp_seq;
3256 unsigned int mtu;
3257 unsigned int off;
3258 int errno;
3259
3260 icmp6 = ip6cp->ip6c_icmp6;
3261 m = ip6cp->ip6c_m;
3262 ip6 = ip6cp->ip6c_ip6;
3263 off = ip6cp->ip6c_off;
3264 dst = &ip6cp->ip6c_finaldst->sin6_addr;
3265
3266 errno = icmp6_errmap(icmp6);
3267 switch (errno) {
3268 case 0:
3269 return;
3270 case EMSGSIZE:
3271 notify = tcp_mtudisc_notify;
3272 break;
3273 case ECONNREFUSED:
3274 if (V_icmp_may_rst)
3275 notify = tcp_drop_syn_sent;
3276 else
3277 notify = tcp_notify;
3278 break;
3279 case EHOSTUNREACH:
3280 /*
3281 * There are only four ICMPs that may reset connection:
3282 * - administratively prohibited
3283 * - port unreachable
3284 * - time exceeded in transit
3285 * - unknown next header
3286 */
3287 if (V_icmp_may_rst &&
3288 ((icmp6->icmp6_type == ICMP6_DST_UNREACH &&
3289 (icmp6->icmp6_code == ICMP6_DST_UNREACH_ADMIN ||
3290 icmp6->icmp6_code == ICMP6_DST_UNREACH_NOPORT)) ||
3291 (icmp6->icmp6_type == ICMP6_TIME_EXCEEDED &&
3292 icmp6->icmp6_code == ICMP6_TIME_EXCEED_TRANSIT) ||
3293 (icmp6->icmp6_type == ICMP6_PARAM_PROB &&
3294 icmp6->icmp6_code == ICMP6_PARAMPROB_NEXTHEADER)))
3295 notify = tcp_drop_syn_sent;
3296 else
3297 notify = tcp_notify;
3298 break;
3299 default:
3300 notify = tcp_notify;
3301 }
3302
3303 /* Check if we can safely get the ports from the tcp hdr */
3304 if (m == NULL ||
3305 (m->m_pkthdr.len <
3306 (int32_t) (off + sizeof(struct tcp_ports)))) {
3307 return;
3308 }
3309 bzero(&t_ports, sizeof(struct tcp_ports));
3310 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
3311 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
3312 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
3313 off += sizeof(struct tcp_ports);
3314 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
3315 goto out;
3316 }
3317 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
3318 if (inp != NULL) {
3319 tp = intotcpcb(inp);
3320 #ifdef TCP_OFFLOAD
3321 if (tp->t_flags & TF_TOE && errno == EMSGSIZE) {
3322 /* MTU discovery for offloaded connections. */
3323 mtu = tcp6_next_pmtu(icmp6);
3324 tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu);
3325 goto out;
3326 }
3327 #endif
3328 if (tp->t_port != port)
3329 goto out;
3330 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
3331 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
3332 if (errno == EMSGSIZE) {
3333 /*
3334 * MTU discovery:
3335 * If we got a needfrag set the MTU
3336 * in the route to the suggested new
3337 * value (if given) and then notify.
3338 */
3339 mtu = tcp6_next_pmtu(icmp6);
3340
3341 bzero(&inc, sizeof(inc));
3342 inc.inc_fibnum = M_GETFIB(m);
3343 inc.inc_flags |= INC_ISIPV6;
3344 inc.inc6_faddr = *dst;
3345 if (in6_setscope(&inc.inc6_faddr,
3346 m->m_pkthdr.rcvif, NULL))
3347 goto out;
3348 /*
3349 * Only process the offered MTU if it
3350 * is smaller than the current one.
3351 */
3352 if (mtu < tp->t_maxseg +
3353 sizeof (struct tcphdr) +
3354 sizeof (struct ip6_hdr)) {
3355 tcp_hc_updatemtu(&inc, mtu);
3356 tcp_mtudisc(inp, mtu);
3357 ICMP6STAT_INC(icp6s_pmtuchg);
3358 }
3359 } else
3360 inp = (*notify)(inp, errno);
3361 }
3362 } else {
3363 bzero(&inc, sizeof(inc));
3364 inc.inc_fibnum = M_GETFIB(m);
3365 inc.inc_flags |= INC_ISIPV6;
3366 inc.inc_fport = t_ports.th_dport;
3367 inc.inc_lport = t_ports.th_sport;
3368 inc.inc6_faddr = *dst;
3369 inc.inc6_laddr = ip6->ip6_src;
3370 syncache_unreach(&inc, icmp_tcp_seq, port);
3371 }
3372 out:
3373 if (inp != NULL)
3374 INP_WUNLOCK(inp);
3375 }
3376
3377 static void
tcp6_ctlinput(struct ip6ctlparam * ctl)3378 tcp6_ctlinput(struct ip6ctlparam *ctl)
3379 {
3380 tcp6_ctlinput_with_port(ctl, htons(0));
3381 }
3382
3383 static void
tcp6_ctlinput_viaudp(udp_tun_icmp_param_t param)3384 tcp6_ctlinput_viaudp(udp_tun_icmp_param_t param)
3385 {
3386 struct ip6ctlparam *ip6cp = param.ip6cp;
3387 struct mbuf *m;
3388 struct udphdr *udp;
3389 uint16_t port;
3390
3391 m = m_pulldown(ip6cp->ip6c_m, ip6cp->ip6c_off, sizeof(struct udphdr), NULL);
3392 if (m == NULL) {
3393 return;
3394 }
3395 udp = mtod(m, struct udphdr *);
3396 if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) {
3397 return;
3398 }
3399 port = udp->uh_dport;
3400 m_adj(m, sizeof(struct udphdr));
3401 if ((m->m_flags & M_PKTHDR) == 0) {
3402 ip6cp->ip6c_m->m_pkthdr.len -= sizeof(struct udphdr);
3403 }
3404 /* Now call in to the normal handling code */
3405 tcp6_ctlinput_with_port(ip6cp, port);
3406 }
3407
3408 #endif /* INET6 */
3409
3410 static uint32_t
tcp_keyed_hash(struct in_conninfo * inc,u_char * key,u_int len)3411 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len)
3412 {
3413 SIPHASH_CTX ctx;
3414 uint32_t hash[2];
3415
3416 KASSERT(len >= SIPHASH_KEY_LENGTH,
3417 ("%s: keylen %u too short ", __func__, len));
3418 SipHash24_Init(&ctx);
3419 SipHash_SetKey(&ctx, (uint8_t *)key);
3420 SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t));
3421 SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t));
3422 switch (inc->inc_flags & INC_ISIPV6) {
3423 #ifdef INET
3424 case 0:
3425 SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr));
3426 SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr));
3427 break;
3428 #endif
3429 #ifdef INET6
3430 case INC_ISIPV6:
3431 SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr));
3432 SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr));
3433 break;
3434 #endif
3435 }
3436 SipHash_Final((uint8_t *)hash, &ctx);
3437
3438 return (hash[0] ^ hash[1]);
3439 }
3440
3441 uint32_t
tcp_new_ts_offset(struct in_conninfo * inc)3442 tcp_new_ts_offset(struct in_conninfo *inc)
3443 {
3444 struct in_conninfo inc_store, *local_inc;
3445
3446 if (!V_tcp_ts_offset_per_conn) {
3447 memcpy(&inc_store, inc, sizeof(struct in_conninfo));
3448 inc_store.inc_lport = 0;
3449 inc_store.inc_fport = 0;
3450 local_inc = &inc_store;
3451 } else {
3452 local_inc = inc;
3453 }
3454 return (tcp_keyed_hash(local_inc, V_ts_offset_secret,
3455 sizeof(V_ts_offset_secret)));
3456 }
3457
3458 /*
3459 * Following is where TCP initial sequence number generation occurs.
3460 *
3461 * There are two places where we must use initial sequence numbers:
3462 * 1. In SYN-ACK packets.
3463 * 2. In SYN packets.
3464 *
3465 * All ISNs for SYN-ACK packets are generated by the syncache. See
3466 * tcp_syncache.c for details.
3467 *
3468 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
3469 * depends on this property. In addition, these ISNs should be
3470 * unguessable so as to prevent connection hijacking. To satisfy
3471 * the requirements of this situation, the algorithm outlined in
3472 * RFC 1948 is used, with only small modifications.
3473 *
3474 * Implementation details:
3475 *
3476 * Time is based off the system timer, and is corrected so that it
3477 * increases by one megabyte per second. This allows for proper
3478 * recycling on high speed LANs while still leaving over an hour
3479 * before rollover.
3480 *
3481 * As reading the *exact* system time is too expensive to be done
3482 * whenever setting up a TCP connection, we increment the time
3483 * offset in two ways. First, a small random positive increment
3484 * is added to isn_offset for each connection that is set up.
3485 * Second, the function tcp_isn_tick fires once per clock tick
3486 * and increments isn_offset as necessary so that sequence numbers
3487 * are incremented at approximately ISN_BYTES_PER_SECOND. The
3488 * random positive increments serve only to ensure that the same
3489 * exact sequence number is never sent out twice (as could otherwise
3490 * happen when a port is recycled in less than the system tick
3491 * interval.)
3492 *
3493 * net.inet.tcp.isn_reseed_interval controls the number of seconds
3494 * between seeding of isn_secret. This is normally set to zero,
3495 * as reseeding should not be necessary.
3496 *
3497 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
3498 * isn_offset_old, and isn_ctx is performed using the ISN lock. In
3499 * general, this means holding an exclusive (write) lock.
3500 */
3501
3502 #define ISN_BYTES_PER_SECOND 1048576
3503 #define ISN_STATIC_INCREMENT 4096
3504 #define ISN_RANDOM_INCREMENT (4096 - 1)
3505 #define ISN_SECRET_LENGTH SIPHASH_KEY_LENGTH
3506
3507 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]);
3508 VNET_DEFINE_STATIC(int, isn_last);
3509 VNET_DEFINE_STATIC(int, isn_last_reseed);
3510 VNET_DEFINE_STATIC(u_int32_t, isn_offset);
3511 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old);
3512
3513 #define V_isn_secret VNET(isn_secret)
3514 #define V_isn_last VNET(isn_last)
3515 #define V_isn_last_reseed VNET(isn_last_reseed)
3516 #define V_isn_offset VNET(isn_offset)
3517 #define V_isn_offset_old VNET(isn_offset_old)
3518
3519 tcp_seq
tcp_new_isn(struct in_conninfo * inc)3520 tcp_new_isn(struct in_conninfo *inc)
3521 {
3522 tcp_seq new_isn;
3523 u_int32_t projected_offset;
3524
3525 ISN_LOCK();
3526 /* Seed if this is the first use, reseed if requested. */
3527 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
3528 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
3529 < (u_int)ticks))) {
3530 arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0);
3531 V_isn_last_reseed = ticks;
3532 }
3533
3534 /* Compute the hash and return the ISN. */
3535 new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret,
3536 sizeof(V_isn_secret));
3537 V_isn_offset += ISN_STATIC_INCREMENT +
3538 (arc4random() & ISN_RANDOM_INCREMENT);
3539 if (ticks != V_isn_last) {
3540 projected_offset = V_isn_offset_old +
3541 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
3542 if (SEQ_GT(projected_offset, V_isn_offset))
3543 V_isn_offset = projected_offset;
3544 V_isn_offset_old = V_isn_offset;
3545 V_isn_last = ticks;
3546 }
3547 new_isn += V_isn_offset;
3548 ISN_UNLOCK();
3549 return (new_isn);
3550 }
3551
3552 /*
3553 * When a specific ICMP unreachable message is received and the
3554 * connection state is SYN-SENT, drop the connection. This behavior
3555 * is controlled by the icmp_may_rst sysctl.
3556 */
3557 static struct inpcb *
tcp_drop_syn_sent(struct inpcb * inp,int errno)3558 tcp_drop_syn_sent(struct inpcb *inp, int errno)
3559 {
3560 struct tcpcb *tp;
3561
3562 NET_EPOCH_ASSERT();
3563 INP_WLOCK_ASSERT(inp);
3564
3565 tp = intotcpcb(inp);
3566 if (tp->t_state != TCPS_SYN_SENT)
3567 return (inp);
3568
3569 if (tp->t_flags & TF_FASTOPEN)
3570 tcp_fastopen_disable_path(tp);
3571
3572 tp = tcp_drop(tp, errno);
3573 if (tp != NULL)
3574 return (inp);
3575 else
3576 return (NULL);
3577 }
3578
3579 /*
3580 * When `need fragmentation' ICMP is received, update our idea of the MSS
3581 * based on the new value. Also nudge TCP to send something, since we
3582 * know the packet we just sent was dropped.
3583 * This duplicates some code in the tcp_mss() function in tcp_input.c.
3584 */
3585 static struct inpcb *
tcp_mtudisc_notify(struct inpcb * inp,int error)3586 tcp_mtudisc_notify(struct inpcb *inp, int error)
3587 {
3588
3589 return (tcp_mtudisc(inp, -1));
3590 }
3591
3592 static struct inpcb *
tcp_mtudisc(struct inpcb * inp,int mtuoffer)3593 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
3594 {
3595 struct tcpcb *tp;
3596 struct socket *so;
3597
3598 INP_WLOCK_ASSERT(inp);
3599
3600 tp = intotcpcb(inp);
3601 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
3602
3603 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
3604
3605 so = inp->inp_socket;
3606 SOCK_SENDBUF_LOCK(so);
3607 /* If the mss is larger than the socket buffer, decrease the mss. */
3608 if (so->so_snd.sb_hiwat < tp->t_maxseg) {
3609 tp->t_maxseg = so->so_snd.sb_hiwat;
3610 if (tp->t_maxseg < V_tcp_mssdflt) {
3611 /*
3612 * The MSS is so small we should not process incoming
3613 * SACK's since we are subject to attack in such a
3614 * case.
3615 */
3616 tp->t_flags2 |= TF2_PROC_SACK_PROHIBIT;
3617 } else {
3618 tp->t_flags2 &= ~TF2_PROC_SACK_PROHIBIT;
3619 }
3620 }
3621 SOCK_SENDBUF_UNLOCK(so);
3622
3623 TCPSTAT_INC(tcps_mturesent);
3624 tp->t_rtttime = 0;
3625 tp->snd_nxt = tp->snd_una;
3626 tcp_free_sackholes(tp);
3627 tp->snd_recover = tp->snd_max;
3628 if (tp->t_flags & TF_SACK_PERMIT)
3629 EXIT_FASTRECOVERY(tp->t_flags);
3630 if (tp->t_fb->tfb_tcp_mtu_chg != NULL) {
3631 /*
3632 * Conceptually the snd_nxt setting
3633 * and freeing sack holes should
3634 * be done by the default stacks
3635 * own tfb_tcp_mtu_chg().
3636 */
3637 tp->t_fb->tfb_tcp_mtu_chg(tp);
3638 }
3639 if (tcp_output(tp) < 0)
3640 return (NULL);
3641 else
3642 return (inp);
3643 }
3644
3645 #ifdef INET
3646 /*
3647 * Look-up the routing entry to the peer of this inpcb. If no route
3648 * is found and it cannot be allocated, then return 0. This routine
3649 * is called by TCP routines that access the rmx structure and by
3650 * tcp_mss_update to get the peer/interface MTU.
3651 */
3652 uint32_t
tcp_maxmtu(struct in_conninfo * inc,struct tcp_ifcap * cap)3653 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
3654 {
3655 struct nhop_object *nh;
3656 struct ifnet *ifp;
3657 uint32_t maxmtu = 0;
3658
3659 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
3660
3661 if (inc->inc_faddr.s_addr != INADDR_ANY) {
3662 nh = fib4_lookup(inc->inc_fibnum, inc->inc_faddr, 0, NHR_NONE, 0);
3663 if (nh == NULL)
3664 return (0);
3665
3666 ifp = nh->nh_ifp;
3667 maxmtu = nh->nh_mtu;
3668
3669 /* Report additional interface capabilities. */
3670 if (cap != NULL) {
3671 if (ifp->if_capenable & IFCAP_TSO4 &&
3672 ifp->if_hwassist & CSUM_TSO) {
3673 cap->ifcap |= CSUM_TSO;
3674 cap->tsomax = ifp->if_hw_tsomax;
3675 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
3676 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
3677 /* XXXKIB IFCAP2_IPSEC_OFFLOAD_TSO */
3678 cap->ipsec_tso = (ifp->if_capenable2 &
3679 IFCAP2_BIT(IFCAP2_IPSEC_OFFLOAD)) != 0;
3680 }
3681 }
3682 }
3683 return (maxmtu);
3684 }
3685 #endif /* INET */
3686
3687 #ifdef INET6
3688 uint32_t
tcp_maxmtu6(struct in_conninfo * inc,struct tcp_ifcap * cap)3689 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
3690 {
3691 struct nhop_object *nh;
3692 struct in6_addr dst6;
3693 uint32_t scopeid;
3694 struct ifnet *ifp;
3695 uint32_t maxmtu = 0;
3696
3697 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
3698
3699 if (inc->inc_flags & INC_IPV6MINMTU)
3700 return (IPV6_MMTU);
3701
3702 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
3703 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
3704 nh = fib6_lookup(inc->inc_fibnum, &dst6, scopeid, NHR_NONE, 0);
3705 if (nh == NULL)
3706 return (0);
3707
3708 ifp = nh->nh_ifp;
3709 maxmtu = nh->nh_mtu;
3710
3711 /* Report additional interface capabilities. */
3712 if (cap != NULL) {
3713 if (ifp->if_capenable & IFCAP_TSO6 &&
3714 ifp->if_hwassist & CSUM_TSO) {
3715 cap->ifcap |= CSUM_TSO;
3716 cap->tsomax = ifp->if_hw_tsomax;
3717 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
3718 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
3719 cap->ipsec_tso = false; /* XXXKIB */
3720 }
3721 }
3722 }
3723
3724 return (maxmtu);
3725 }
3726
3727 /*
3728 * Handle setsockopt(IPV6_USE_MIN_MTU) by a TCP stack.
3729 *
3730 * XXXGL: we are updating inpcb here with INC_IPV6MINMTU flag.
3731 * The right place to do that is ip6_setpktopt() that has just been
3732 * executed. By the way it just filled ip6po_minmtu for us.
3733 */
3734 void
tcp6_use_min_mtu(struct tcpcb * tp)3735 tcp6_use_min_mtu(struct tcpcb *tp)
3736 {
3737 struct inpcb *inp = tptoinpcb(tp);
3738
3739 INP_WLOCK_ASSERT(inp);
3740 /*
3741 * In case of the IPV6_USE_MIN_MTU socket
3742 * option, the INC_IPV6MINMTU flag to announce
3743 * a corresponding MSS during the initial
3744 * handshake. If the TCP connection is not in
3745 * the front states, just reduce the MSS being
3746 * used. This avoids the sending of TCP
3747 * segments which will be fragmented at the
3748 * IPv6 layer.
3749 */
3750 inp->inp_inc.inc_flags |= INC_IPV6MINMTU;
3751 if ((tp->t_state >= TCPS_SYN_SENT) &&
3752 (inp->inp_inc.inc_flags & INC_ISIPV6)) {
3753 struct ip6_pktopts *opt;
3754
3755 opt = inp->in6p_outputopts;
3756 if (opt != NULL && opt->ip6po_minmtu == IP6PO_MINMTU_ALL &&
3757 tp->t_maxseg > TCP6_MSS) {
3758 tp->t_maxseg = TCP6_MSS;
3759 if (tp->t_maxseg < V_tcp_mssdflt) {
3760 /*
3761 * The MSS is so small we should not process incoming
3762 * SACK's since we are subject to attack in such a
3763 * case.
3764 */
3765 tp->t_flags2 |= TF2_PROC_SACK_PROHIBIT;
3766 } else {
3767 tp->t_flags2 &= ~TF2_PROC_SACK_PROHIBIT;
3768 }
3769 }
3770 }
3771 }
3772 #endif /* INET6 */
3773
3774 /*
3775 * Calculate effective SMSS per RFC5681 definition for a given TCP
3776 * connection at its current state, taking into account SACK and etc.
3777 */
3778 u_int
tcp_maxseg(const struct tcpcb * tp)3779 tcp_maxseg(const struct tcpcb *tp)
3780 {
3781 u_int optlen;
3782
3783 if (tp->t_flags & TF_NOOPT)
3784 return (tp->t_maxseg);
3785
3786 /*
3787 * Here we have a simplified code from tcp_addoptions(),
3788 * without a proper loop, and having most of paddings hardcoded.
3789 * We might make mistakes with padding here in some edge cases,
3790 * but this is harmless, since result of tcp_maxseg() is used
3791 * only in cwnd and ssthresh estimations.
3792 */
3793 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3794 if (tp->t_flags & TF_RCVD_TSTMP)
3795 optlen = TCPOLEN_TSTAMP_APPA;
3796 else
3797 optlen = 0;
3798 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3799 if (tp->t_flags & TF_SIGNATURE)
3800 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE);
3801 #endif
3802 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
3803 optlen += TCPOLEN_SACKHDR;
3804 optlen += tp->rcv_numsacks * TCPOLEN_SACK;
3805 optlen = PADTCPOLEN(optlen);
3806 }
3807 } else {
3808 if (tp->t_flags & TF_REQ_TSTMP)
3809 optlen = TCPOLEN_TSTAMP_APPA;
3810 else
3811 optlen = PADTCPOLEN(TCPOLEN_MAXSEG);
3812 if (tp->t_flags & TF_REQ_SCALE)
3813 optlen += PADTCPOLEN(TCPOLEN_WINDOW);
3814 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3815 if (tp->t_flags & TF_SIGNATURE)
3816 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE);
3817 #endif
3818 if (tp->t_flags & TF_SACK_PERMIT)
3819 optlen += PADTCPOLEN(TCPOLEN_SACK_PERMITTED);
3820 }
3821 optlen = min(optlen, TCP_MAXOLEN);
3822 return (tp->t_maxseg - optlen);
3823 }
3824
3825
3826 u_int
tcp_fixed_maxseg(const struct tcpcb * tp)3827 tcp_fixed_maxseg(const struct tcpcb *tp)
3828 {
3829 int optlen;
3830
3831 if (tp->t_flags & TF_NOOPT)
3832 return (tp->t_maxseg);
3833
3834 /*
3835 * Here we have a simplified code from tcp_addoptions(),
3836 * without a proper loop, and having most of paddings hardcoded.
3837 * We only consider fixed options that we would send every
3838 * time I.e. SACK is not considered. This is important
3839 * for cc modules to figure out what the modulo of the
3840 * cwnd should be.
3841 */
3842 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3843 if (tp->t_flags & TF_RCVD_TSTMP)
3844 optlen = TCPOLEN_TSTAMP_APPA;
3845 else
3846 optlen = 0;
3847 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3848 if (tp->t_flags & TF_SIGNATURE)
3849 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE);
3850 #endif
3851 } else {
3852 if (tp->t_flags & TF_REQ_TSTMP)
3853 optlen = TCPOLEN_TSTAMP_APPA;
3854 else
3855 optlen = PADTCPOLEN(TCPOLEN_MAXSEG);
3856 if (tp->t_flags & TF_REQ_SCALE)
3857 optlen += PADTCPOLEN(TCPOLEN_WINDOW);
3858 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3859 if (tp->t_flags & TF_SIGNATURE)
3860 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE);
3861 #endif
3862 if (tp->t_flags & TF_SACK_PERMIT)
3863 optlen += PADTCPOLEN(TCPOLEN_SACK_PERMITTED);
3864 }
3865 optlen = min(optlen, TCP_MAXOLEN);
3866 return (tp->t_maxseg - optlen);
3867 }
3868
3869
3870
3871 static int
sysctl_drop(SYSCTL_HANDLER_ARGS)3872 sysctl_drop(SYSCTL_HANDLER_ARGS)
3873 {
3874 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
3875 struct sockaddr_storage addrs[2];
3876 struct inpcb *inp;
3877 struct tcpcb *tp;
3878 #ifdef INET
3879 struct sockaddr_in *fin = NULL, *lin = NULL;
3880 #endif
3881 struct epoch_tracker et;
3882 #ifdef INET6
3883 struct sockaddr_in6 *fin6, *lin6;
3884 #endif
3885 int error;
3886
3887 inp = NULL;
3888 #ifdef INET6
3889 fin6 = lin6 = NULL;
3890 #endif
3891 error = 0;
3892
3893 if (req->oldptr != NULL || req->oldlen != 0)
3894 return (EINVAL);
3895 if (req->newptr == NULL)
3896 return (EPERM);
3897 if (req->newlen < sizeof(addrs))
3898 return (ENOMEM);
3899 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3900 if (error)
3901 return (error);
3902
3903 switch (addrs[0].ss_family) {
3904 #ifdef INET6
3905 case AF_INET6:
3906 fin6 = (struct sockaddr_in6 *)&addrs[0];
3907 lin6 = (struct sockaddr_in6 *)&addrs[1];
3908 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3909 lin6->sin6_len != sizeof(struct sockaddr_in6))
3910 return (EINVAL);
3911 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3912 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3913 return (EINVAL);
3914 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3915 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3916 #ifdef INET
3917 fin = (struct sockaddr_in *)&addrs[0];
3918 lin = (struct sockaddr_in *)&addrs[1];
3919 #endif
3920 break;
3921 }
3922 error = sa6_embedscope(fin6, V_ip6_use_defzone);
3923 if (error)
3924 return (error);
3925 error = sa6_embedscope(lin6, V_ip6_use_defzone);
3926 if (error)
3927 return (error);
3928 break;
3929 #endif
3930 #ifdef INET
3931 case AF_INET:
3932 fin = (struct sockaddr_in *)&addrs[0];
3933 lin = (struct sockaddr_in *)&addrs[1];
3934 if (fin->sin_len != sizeof(struct sockaddr_in) ||
3935 lin->sin_len != sizeof(struct sockaddr_in))
3936 return (EINVAL);
3937 break;
3938 #endif
3939 default:
3940 return (EINVAL);
3941 }
3942 NET_EPOCH_ENTER(et);
3943 switch (addrs[0].ss_family) {
3944 #ifdef INET6
3945 case AF_INET6:
3946 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3947 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3948 INPLOOKUP_WLOCKPCB, NULL);
3949 break;
3950 #endif
3951 #ifdef INET
3952 case AF_INET:
3953 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3954 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3955 break;
3956 #endif
3957 }
3958 if (inp != NULL) {
3959 if (!SOLISTENING(inp->inp_socket)) {
3960 tp = intotcpcb(inp);
3961 tp = tcp_drop(tp, ECONNABORTED);
3962 if (tp != NULL)
3963 INP_WUNLOCK(inp);
3964 } else
3965 INP_WUNLOCK(inp);
3966 } else
3967 error = ESRCH;
3968 NET_EPOCH_EXIT(et);
3969 return (error);
3970 }
3971
3972 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
3973 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3974 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_drop, "",
3975 "Drop TCP connection");
3976
3977 static int
tcp_sysctl_setsockopt(SYSCTL_HANDLER_ARGS)3978 tcp_sysctl_setsockopt(SYSCTL_HANDLER_ARGS)
3979 {
3980 return (sysctl_setsockopt(oidp, arg1, arg2, req, &V_tcbinfo,
3981 &tcp_ctloutput_set));
3982 }
3983
3984 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, setsockopt,
3985 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3986 CTLFLAG_MPSAFE, NULL, 0, tcp_sysctl_setsockopt, "",
3987 "Set socket option for TCP endpoint");
3988
3989 #ifdef KERN_TLS
3990 static int
sysctl_switch_tls(SYSCTL_HANDLER_ARGS)3991 sysctl_switch_tls(SYSCTL_HANDLER_ARGS)
3992 {
3993 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
3994 struct sockaddr_storage addrs[2];
3995 struct inpcb *inp;
3996 #ifdef INET
3997 struct sockaddr_in *fin = NULL, *lin = NULL;
3998 #endif
3999 struct epoch_tracker et;
4000 #ifdef INET6
4001 struct sockaddr_in6 *fin6, *lin6;
4002 #endif
4003 int error;
4004
4005 inp = NULL;
4006 #ifdef INET6
4007 fin6 = lin6 = NULL;
4008 #endif
4009 error = 0;
4010
4011 if (req->oldptr != NULL || req->oldlen != 0)
4012 return (EINVAL);
4013 if (req->newptr == NULL)
4014 return (EPERM);
4015 if (req->newlen < sizeof(addrs))
4016 return (ENOMEM);
4017 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
4018 if (error)
4019 return (error);
4020
4021 switch (addrs[0].ss_family) {
4022 #ifdef INET6
4023 case AF_INET6:
4024 fin6 = (struct sockaddr_in6 *)&addrs[0];
4025 lin6 = (struct sockaddr_in6 *)&addrs[1];
4026 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
4027 lin6->sin6_len != sizeof(struct sockaddr_in6))
4028 return (EINVAL);
4029 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
4030 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
4031 return (EINVAL);
4032 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
4033 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
4034 #ifdef INET
4035 fin = (struct sockaddr_in *)&addrs[0];
4036 lin = (struct sockaddr_in *)&addrs[1];
4037 #endif
4038 break;
4039 }
4040 error = sa6_embedscope(fin6, V_ip6_use_defzone);
4041 if (error)
4042 return (error);
4043 error = sa6_embedscope(lin6, V_ip6_use_defzone);
4044 if (error)
4045 return (error);
4046 break;
4047 #endif
4048 #ifdef INET
4049 case AF_INET:
4050 fin = (struct sockaddr_in *)&addrs[0];
4051 lin = (struct sockaddr_in *)&addrs[1];
4052 if (fin->sin_len != sizeof(struct sockaddr_in) ||
4053 lin->sin_len != sizeof(struct sockaddr_in))
4054 return (EINVAL);
4055 break;
4056 #endif
4057 default:
4058 return (EINVAL);
4059 }
4060 NET_EPOCH_ENTER(et);
4061 switch (addrs[0].ss_family) {
4062 #ifdef INET6
4063 case AF_INET6:
4064 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
4065 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
4066 INPLOOKUP_WLOCKPCB, NULL);
4067 break;
4068 #endif
4069 #ifdef INET
4070 case AF_INET:
4071 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
4072 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
4073 break;
4074 #endif
4075 }
4076 NET_EPOCH_EXIT(et);
4077 if (inp != NULL) {
4078 struct socket *so;
4079
4080 so = inp->inp_socket;
4081 soref(so);
4082 error = ktls_set_tx_mode(so,
4083 arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET);
4084 INP_WUNLOCK(inp);
4085 sorele(so);
4086 } else
4087 error = ESRCH;
4088 return (error);
4089 }
4090
4091 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls,
4092 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
4093 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_switch_tls, "",
4094 "Switch TCP connection to SW TLS");
4095 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls,
4096 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
4097 CTLFLAG_NEEDGIANT, NULL, 1, sysctl_switch_tls, "",
4098 "Switch TCP connection to ifnet TLS");
4099 #endif
4100
4101 /*
4102 * Generate a standardized TCP log line for use throughout the
4103 * tcp subsystem. Memory allocation is done with M_NOWAIT to
4104 * allow use in the interrupt context.
4105 *
4106 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
4107 * NB: The function may return NULL if memory allocation failed.
4108 *
4109 * Due to header inclusion and ordering limitations the struct ip
4110 * and ip6_hdr pointers have to be passed as void pointers.
4111 */
4112 char *
tcp_log_vain(struct in_conninfo * inc,struct tcphdr * th,const void * ip4hdr,const void * ip6hdr)4113 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr,
4114 const void *ip6hdr)
4115 {
4116
4117 /* Is logging enabled? */
4118 if (V_tcp_log_in_vain == 0)
4119 return (NULL);
4120
4121 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
4122 }
4123
4124 char *
tcp_log_addrs(struct in_conninfo * inc,struct tcphdr * th,const void * ip4hdr,const void * ip6hdr)4125 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr,
4126 const void *ip6hdr)
4127 {
4128
4129 /* Is logging enabled? */
4130 if (tcp_log_debug == 0)
4131 return (NULL);
4132
4133 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
4134 }
4135
4136 static char *
tcp_log_addr(struct in_conninfo * inc,struct tcphdr * th,const void * ip4hdr,const void * ip6hdr)4137 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr,
4138 const void *ip6hdr)
4139 {
4140 char *s, *sp;
4141 size_t size;
4142 #ifdef INET
4143 const struct ip *ip = (const struct ip *)ip4hdr;
4144 #endif
4145 #ifdef INET6
4146 const struct ip6_hdr *ip6 = (const struct ip6_hdr *)ip6hdr;
4147 #endif /* INET6 */
4148
4149 /*
4150 * The log line looks like this:
4151 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
4152 */
4153 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
4154 sizeof(PRINT_TH_FLAGS) + 1 +
4155 #ifdef INET6
4156 2 * INET6_ADDRSTRLEN;
4157 #else
4158 2 * INET_ADDRSTRLEN;
4159 #endif /* INET6 */
4160
4161 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
4162 if (s == NULL)
4163 return (NULL);
4164
4165 strcat(s, "TCP: [");
4166 sp = s + strlen(s);
4167
4168 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
4169 inet_ntoa_r(inc->inc_faddr, sp);
4170 sp = s + strlen(s);
4171 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
4172 sp = s + strlen(s);
4173 inet_ntoa_r(inc->inc_laddr, sp);
4174 sp = s + strlen(s);
4175 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
4176 #ifdef INET6
4177 } else if (inc) {
4178 ip6_sprintf(sp, &inc->inc6_faddr);
4179 sp = s + strlen(s);
4180 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
4181 sp = s + strlen(s);
4182 ip6_sprintf(sp, &inc->inc6_laddr);
4183 sp = s + strlen(s);
4184 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
4185 } else if (ip6 && th) {
4186 ip6_sprintf(sp, &ip6->ip6_src);
4187 sp = s + strlen(s);
4188 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
4189 sp = s + strlen(s);
4190 ip6_sprintf(sp, &ip6->ip6_dst);
4191 sp = s + strlen(s);
4192 sprintf(sp, "]:%i", ntohs(th->th_dport));
4193 #endif /* INET6 */
4194 #ifdef INET
4195 } else if (ip && th) {
4196 inet_ntoa_r(ip->ip_src, sp);
4197 sp = s + strlen(s);
4198 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
4199 sp = s + strlen(s);
4200 inet_ntoa_r(ip->ip_dst, sp);
4201 sp = s + strlen(s);
4202 sprintf(sp, "]:%i", ntohs(th->th_dport));
4203 #endif /* INET */
4204 } else {
4205 free(s, M_TCPLOG);
4206 return (NULL);
4207 }
4208 sp = s + strlen(s);
4209 if (th)
4210 sprintf(sp, " tcpflags 0x%b", tcp_get_flags(th), PRINT_TH_FLAGS);
4211 if (*(s + size - 1) != '\0')
4212 panic("%s: string too long", __func__);
4213 return (s);
4214 }
4215
4216 /*
4217 * A subroutine which makes it easy to track TCP state changes with DTrace.
4218 * This function shouldn't be called for t_state initializations that don't
4219 * correspond to actual TCP state transitions.
4220 */
4221 void
tcp_state_change(struct tcpcb * tp,int newstate)4222 tcp_state_change(struct tcpcb *tp, int newstate)
4223 {
4224 #if defined(KDTRACE_HOOKS)
4225 int pstate = tp->t_state;
4226 #endif
4227
4228 TCPSTATES_DEC(tp->t_state);
4229 TCPSTATES_INC(newstate);
4230 tp->t_state = newstate;
4231 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
4232 }
4233
4234 /*
4235 * Create an external-format (``xtcpcb'') structure using the information in
4236 * the kernel-format tcpcb structure pointed to by tp. This is done to
4237 * reduce the spew of irrelevant information over this interface, to isolate
4238 * user code from changes in the kernel structure, and potentially to provide
4239 * information-hiding if we decide that some of this information should be
4240 * hidden from users.
4241 */
4242 void
tcp_inptoxtp(const struct inpcb * inp,struct xtcpcb * xt)4243 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
4244 {
4245 struct tcpcb *tp = intotcpcb(inp);
4246 sbintime_t now;
4247
4248 bzero(xt, sizeof(*xt));
4249 xt->t_state = tp->t_state;
4250 xt->t_logstate = tcp_get_bblog_state(tp);
4251 xt->t_flags = tp->t_flags;
4252 xt->t_sndzerowin = tp->t_sndzerowin;
4253 xt->t_sndrexmitpack = tp->t_sndrexmitpack;
4254 xt->t_rcvoopack = tp->t_rcvoopack;
4255 xt->t_rcv_wnd = tp->rcv_wnd;
4256 xt->t_snd_wnd = tp->snd_wnd;
4257 xt->t_snd_cwnd = tp->snd_cwnd;
4258 xt->t_snd_ssthresh = tp->snd_ssthresh;
4259 xt->t_dsack_bytes = tp->t_dsack_bytes;
4260 xt->t_dsack_tlp_bytes = tp->t_dsack_tlp_bytes;
4261 xt->t_dsack_pack = tp->t_dsack_pack;
4262 xt->t_maxseg = tp->t_maxseg;
4263 xt->xt_ecn = (tp->t_flags2 & TF2_ECN_PERMIT) ? 1 : 0 +
4264 (tp->t_flags2 & TF2_ACE_PERMIT) ? 2 : 0;
4265
4266 now = getsbinuptime();
4267 #define COPYTIMER(which,where) do { \
4268 if (tp->t_timers[which] != SBT_MAX) \
4269 xt->where = (tp->t_timers[which] - now) / SBT_1MS; \
4270 else \
4271 xt->where = 0; \
4272 } while (0)
4273 COPYTIMER(TT_DELACK, tt_delack);
4274 COPYTIMER(TT_REXMT, tt_rexmt);
4275 COPYTIMER(TT_PERSIST, tt_persist);
4276 COPYTIMER(TT_KEEP, tt_keep);
4277 COPYTIMER(TT_2MSL, tt_2msl);
4278 #undef COPYTIMER
4279 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
4280
4281 xt->xt_encaps_port = tp->t_port;
4282 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
4283 TCP_FUNCTION_NAME_LEN_MAX);
4284 bcopy(CC_ALGO(tp)->name, xt->xt_cc, TCP_CA_NAME_MAX);
4285 #ifdef TCP_BLACKBOX
4286 (void)tcp_log_get_id(tp, xt->xt_logid);
4287 #endif
4288
4289 xt->xt_len = sizeof(struct xtcpcb);
4290 in_pcbtoxinpcb(inp, &xt->xt_inp);
4291 }
4292
4293 void
tcp_log_end_status(struct tcpcb * tp,uint8_t status)4294 tcp_log_end_status(struct tcpcb *tp, uint8_t status)
4295 {
4296 uint32_t bit, i;
4297
4298 if ((tp == NULL) ||
4299 (status > TCP_EI_STATUS_MAX_VALUE) ||
4300 (status == 0)) {
4301 /* Invalid */
4302 return;
4303 }
4304 if (status > (sizeof(uint32_t) * 8)) {
4305 /* Should this be a KASSERT? */
4306 return;
4307 }
4308 bit = 1U << (status - 1);
4309 if (bit & tp->t_end_info_status) {
4310 /* already logged */
4311 return;
4312 }
4313 for (i = 0; i < TCP_END_BYTE_INFO; i++) {
4314 if (tp->t_end_info_bytes[i] == TCP_EI_EMPTY_SLOT) {
4315 tp->t_end_info_bytes[i] = status;
4316 tp->t_end_info_status |= bit;
4317 break;
4318 }
4319 }
4320 }
4321
4322 int
tcp_can_enable_pacing(void)4323 tcp_can_enable_pacing(void)
4324 {
4325
4326 if ((tcp_pacing_limit == -1) ||
4327 (tcp_pacing_limit > number_of_tcp_connections_pacing)) {
4328 atomic_fetchadd_int(&number_of_tcp_connections_pacing, 1);
4329 shadow_num_connections = number_of_tcp_connections_pacing;
4330 return (1);
4331 } else {
4332 counter_u64_add(tcp_pacing_failures, 1);
4333 return (0);
4334 }
4335 }
4336
4337 int
tcp_incr_dgp_pacing_cnt(void)4338 tcp_incr_dgp_pacing_cnt(void)
4339 {
4340 if ((tcp_dgp_limit == -1) ||
4341 (tcp_dgp_limit > number_of_dgp_connections)) {
4342 atomic_fetchadd_int(&number_of_dgp_connections, 1);
4343 shadow_tcp_pacing_dgp = number_of_dgp_connections;
4344 return (1);
4345 } else {
4346 counter_u64_add(tcp_dgp_failures, 1);
4347 return (0);
4348 }
4349 }
4350
4351 static uint8_t tcp_dgp_warning = 0;
4352
4353 void
tcp_dec_dgp_pacing_cnt(void)4354 tcp_dec_dgp_pacing_cnt(void)
4355 {
4356 uint32_t ret;
4357
4358 ret = atomic_fetchadd_int(&number_of_dgp_connections, -1);
4359 shadow_tcp_pacing_dgp = number_of_dgp_connections;
4360 KASSERT(ret != 0, ("number_of_dgp_connections -1 would cause wrap?"));
4361 if (ret == 0) {
4362 if (tcp_dgp_limit != -1) {
4363 printf("Warning all DGP is now disabled, count decrements invalidly!\n");
4364 tcp_dgp_limit = 0;
4365 tcp_dgp_warning = 1;
4366 } else if (tcp_dgp_warning == 0) {
4367 printf("Warning DGP pacing is invalid, invalid decrement\n");
4368 tcp_dgp_warning = 1;
4369 }
4370 }
4371
4372 }
4373
4374 static uint8_t tcp_pacing_warning = 0;
4375
4376 void
tcp_decrement_paced_conn(void)4377 tcp_decrement_paced_conn(void)
4378 {
4379 uint32_t ret;
4380
4381 ret = atomic_fetchadd_int(&number_of_tcp_connections_pacing, -1);
4382 shadow_num_connections = number_of_tcp_connections_pacing;
4383 KASSERT(ret != 0, ("tcp_paced_connection_exits -1 would cause wrap?"));
4384 if (ret == 0) {
4385 if (tcp_pacing_limit != -1) {
4386 printf("Warning all pacing is now disabled, count decrements invalidly!\n");
4387 tcp_pacing_limit = 0;
4388 } else if (tcp_pacing_warning == 0) {
4389 printf("Warning pacing count is invalid, invalid decrement\n");
4390 tcp_pacing_warning = 1;
4391 }
4392 }
4393 }
4394
4395 static void
tcp_default_switch_failed(struct tcpcb * tp)4396 tcp_default_switch_failed(struct tcpcb *tp)
4397 {
4398 /*
4399 * If a switch fails we only need to
4400 * care about two things:
4401 * a) The t_flags2
4402 * and
4403 * b) The timer granularity.
4404 * Timeouts, at least for now, don't use the
4405 * old callout system in the other stacks so
4406 * those are hopefully safe.
4407 */
4408 tcp_lro_features_off(tp);
4409 tcp_change_time_units(tp, TCP_TMR_GRANULARITY_TICKS);
4410 }
4411
4412 #ifdef TCP_ACCOUNTING
4413 int
tcp_do_ack_accounting(struct tcpcb * tp,struct tcphdr * th,struct tcpopt * to,uint32_t tiwin,int mss)4414 tcp_do_ack_accounting(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to, uint32_t tiwin, int mss)
4415 {
4416 if (SEQ_LT(th->th_ack, tp->snd_una)) {
4417 /* Do we have a SACK? */
4418 if (to->to_flags & TOF_SACK) {
4419 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
4420 tp->tcp_cnt_counters[ACK_SACK]++;
4421 }
4422 return (ACK_SACK);
4423 } else {
4424 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
4425 tp->tcp_cnt_counters[ACK_BEHIND]++;
4426 }
4427 return (ACK_BEHIND);
4428 }
4429 } else if (th->th_ack == tp->snd_una) {
4430 /* Do we have a SACK? */
4431 if (to->to_flags & TOF_SACK) {
4432 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
4433 tp->tcp_cnt_counters[ACK_SACK]++;
4434 }
4435 return (ACK_SACK);
4436 } else if (tiwin != tp->snd_wnd) {
4437 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
4438 tp->tcp_cnt_counters[ACK_RWND]++;
4439 }
4440 return (ACK_RWND);
4441 } else {
4442 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
4443 tp->tcp_cnt_counters[ACK_DUPACK]++;
4444 }
4445 return (ACK_DUPACK);
4446 }
4447 } else {
4448 if (!SEQ_GT(th->th_ack, tp->snd_max)) {
4449 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
4450 tp->tcp_cnt_counters[CNT_OF_ACKS_IN] += (((th->th_ack - tp->snd_una) + mss - 1)/mss);
4451 }
4452 }
4453 if (to->to_flags & TOF_SACK) {
4454 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
4455 tp->tcp_cnt_counters[ACK_CUMACK_SACK]++;
4456 }
4457 return (ACK_CUMACK_SACK);
4458 } else {
4459 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
4460 tp->tcp_cnt_counters[ACK_CUMACK]++;
4461 }
4462 return (ACK_CUMACK);
4463 }
4464 }
4465 }
4466 #endif
4467
4468 void
tcp_change_time_units(struct tcpcb * tp,int granularity)4469 tcp_change_time_units(struct tcpcb *tp, int granularity)
4470 {
4471 if (tp->t_tmr_granularity == granularity) {
4472 /* We are there */
4473 return;
4474 }
4475 if (granularity == TCP_TMR_GRANULARITY_USEC) {
4476 KASSERT((tp->t_tmr_granularity == TCP_TMR_GRANULARITY_TICKS),
4477 ("Granularity is not TICKS its %u in tp:%p",
4478 tp->t_tmr_granularity, tp));
4479 tp->t_rttlow = TICKS_2_USEC(tp->t_rttlow);
4480 if (tp->t_srtt > 1) {
4481 uint32_t val, frac;
4482
4483 val = tp->t_srtt >> TCP_RTT_SHIFT;
4484 frac = tp->t_srtt & 0x1f;
4485 tp->t_srtt = TICKS_2_USEC(val);
4486 /*
4487 * frac is the fractional part of the srtt (if any)
4488 * but its in ticks and every bit represents
4489 * 1/32nd of a hz.
4490 */
4491 if (frac) {
4492 if (hz == 1000) {
4493 frac = (((uint64_t)frac * (uint64_t)HPTS_USEC_IN_MSEC) / (uint64_t)TCP_RTT_SCALE);
4494 } else {
4495 frac = (((uint64_t)frac * (uint64_t)HPTS_USEC_IN_SEC) / ((uint64_t)(hz) * (uint64_t)TCP_RTT_SCALE));
4496 }
4497 tp->t_srtt += frac;
4498 }
4499 }
4500 if (tp->t_rttvar) {
4501 uint32_t val, frac;
4502
4503 val = tp->t_rttvar >> TCP_RTTVAR_SHIFT;
4504 frac = tp->t_rttvar & 0x1f;
4505 tp->t_rttvar = TICKS_2_USEC(val);
4506 /*
4507 * frac is the fractional part of the srtt (if any)
4508 * but its in ticks and every bit represents
4509 * 1/32nd of a hz.
4510 */
4511 if (frac) {
4512 if (hz == 1000) {
4513 frac = (((uint64_t)frac * (uint64_t)HPTS_USEC_IN_MSEC) / (uint64_t)TCP_RTT_SCALE);
4514 } else {
4515 frac = (((uint64_t)frac * (uint64_t)HPTS_USEC_IN_SEC) / ((uint64_t)(hz) * (uint64_t)TCP_RTT_SCALE));
4516 }
4517 tp->t_rttvar += frac;
4518 }
4519 }
4520 tp->t_tmr_granularity = TCP_TMR_GRANULARITY_USEC;
4521 } else if (granularity == TCP_TMR_GRANULARITY_TICKS) {
4522 /* Convert back to ticks, with */
4523 KASSERT((tp->t_tmr_granularity == TCP_TMR_GRANULARITY_USEC),
4524 ("Granularity is not USEC its %u in tp:%p",
4525 tp->t_tmr_granularity, tp));
4526 if (tp->t_srtt > 1) {
4527 uint32_t val, frac;
4528
4529 val = USEC_2_TICKS(tp->t_srtt);
4530 frac = tp->t_srtt % (HPTS_USEC_IN_SEC / hz);
4531 tp->t_srtt = val << TCP_RTT_SHIFT;
4532 /*
4533 * frac is the fractional part here is left
4534 * over from converting to hz and shifting.
4535 * We need to convert this to the 5 bit
4536 * remainder.
4537 */
4538 if (frac) {
4539 if (hz == 1000) {
4540 frac = (((uint64_t)frac * (uint64_t)TCP_RTT_SCALE) / (uint64_t)HPTS_USEC_IN_MSEC);
4541 } else {
4542 frac = (((uint64_t)frac * (uint64_t)(hz) * (uint64_t)TCP_RTT_SCALE) /(uint64_t)HPTS_USEC_IN_SEC);
4543 }
4544 tp->t_srtt += frac;
4545 }
4546 }
4547 if (tp->t_rttvar) {
4548 uint32_t val, frac;
4549
4550 val = USEC_2_TICKS(tp->t_rttvar);
4551 frac = tp->t_rttvar % (HPTS_USEC_IN_SEC / hz);
4552 tp->t_rttvar = val << TCP_RTTVAR_SHIFT;
4553 /*
4554 * frac is the fractional part here is left
4555 * over from converting to hz and shifting.
4556 * We need to convert this to the 4 bit
4557 * remainder.
4558 */
4559 if (frac) {
4560 if (hz == 1000) {
4561 frac = (((uint64_t)frac * (uint64_t)TCP_RTTVAR_SCALE) / (uint64_t)HPTS_USEC_IN_MSEC);
4562 } else {
4563 frac = (((uint64_t)frac * (uint64_t)(hz) * (uint64_t)TCP_RTTVAR_SCALE) /(uint64_t)HPTS_USEC_IN_SEC);
4564 }
4565 tp->t_rttvar += frac;
4566 }
4567 }
4568 tp->t_rttlow = USEC_2_TICKS(tp->t_rttlow);
4569 tp->t_tmr_granularity = TCP_TMR_GRANULARITY_TICKS;
4570 }
4571 #ifdef INVARIANTS
4572 else {
4573 panic("Unknown granularity:%d tp:%p",
4574 granularity, tp);
4575 }
4576 #endif
4577 }
4578
4579 void
tcp_handle_orphaned_packets(struct tcpcb * tp)4580 tcp_handle_orphaned_packets(struct tcpcb *tp)
4581 {
4582 struct mbuf *save, *m, *prev;
4583 /*
4584 * Called when a stack switch is occuring from the fini()
4585 * of the old stack. We assue the init() as already been
4586 * run of the new stack and it has set the t_flags2 to
4587 * what it supports. This function will then deal with any
4588 * differences i.e. cleanup packets that maybe queued that
4589 * the newstack does not support.
4590 */
4591
4592 if (tp->t_flags2 & TF2_MBUF_L_ACKS)
4593 return;
4594 if ((tp->t_flags2 & TF2_SUPPORTS_MBUFQ) == 0 &&
4595 !STAILQ_EMPTY(&tp->t_inqueue)) {
4596 /*
4597 * It is unsafe to process the packets since a
4598 * reset may be lurking in them (its rare but it
4599 * can occur). If we were to find a RST, then we
4600 * would end up dropping the connection and the
4601 * INP lock, so when we return the caller (tcp_usrreq)
4602 * will blow up when it trys to unlock the inp.
4603 * This new stack does not do any fancy LRO features
4604 * so all we can do is toss the packets.
4605 */
4606 m = STAILQ_FIRST(&tp->t_inqueue);
4607 STAILQ_INIT(&tp->t_inqueue);
4608 STAILQ_FOREACH_FROM_SAFE(m, &tp->t_inqueue, m_stailqpkt, save)
4609 m_freem(m);
4610 } else {
4611 /*
4612 * Here we have a stack that does mbuf queuing but
4613 * does not support compressed ack's. We must
4614 * walk all the mbufs and discard any compressed acks.
4615 */
4616 STAILQ_FOREACH_SAFE(m, &tp->t_inqueue, m_stailqpkt, save) {
4617 if (m->m_flags & M_ACKCMP) {
4618 if (m == STAILQ_FIRST(&tp->t_inqueue))
4619 STAILQ_REMOVE_HEAD(&tp->t_inqueue,
4620 m_stailqpkt);
4621 else
4622 STAILQ_REMOVE_AFTER(&tp->t_inqueue,
4623 prev, m_stailqpkt);
4624 m_freem(m);
4625 } else
4626 prev = m;
4627 }
4628 }
4629 }
4630
4631 #ifdef TCP_REQUEST_TRK
4632 uint32_t
tcp_estimate_tls_overhead(struct socket * so,uint64_t tls_usr_bytes)4633 tcp_estimate_tls_overhead(struct socket *so, uint64_t tls_usr_bytes)
4634 {
4635 #ifdef KERN_TLS
4636 struct ktls_session *tls;
4637 uint32_t rec_oh, records;
4638
4639 tls = so->so_snd.sb_tls_info;
4640 if (tls == NULL)
4641 return (0);
4642
4643 rec_oh = tls->params.tls_hlen + tls->params.tls_tlen;
4644 records = ((tls_usr_bytes + tls->params.max_frame_len - 1)/tls->params.max_frame_len);
4645 return (records * rec_oh);
4646 #else
4647 return (0);
4648 #endif
4649 }
4650
4651 extern uint32_t tcp_stale_entry_time;
4652 uint32_t tcp_stale_entry_time = 250000;
4653 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, usrlog_stale, CTLFLAG_RW,
4654 &tcp_stale_entry_time, 250000, "Time that a tcpreq entry without a sendfile ages out");
4655
4656 void
tcp_req_log_req_info(struct tcpcb * tp,struct tcp_sendfile_track * req,uint16_t slot,uint8_t val,uint64_t offset,uint64_t nbytes)4657 tcp_req_log_req_info(struct tcpcb *tp, struct tcp_sendfile_track *req,
4658 uint16_t slot, uint8_t val, uint64_t offset, uint64_t nbytes)
4659 {
4660 if (tcp_bblogging_on(tp)) {
4661 union tcp_log_stackspecific log;
4662 struct timeval tv;
4663
4664 memset(&log, 0, sizeof(log));
4665 log.u_bbr.inhpts = tcp_in_hpts(tp);
4666 log.u_bbr.flex8 = val;
4667 log.u_bbr.rttProp = req->timestamp;
4668 log.u_bbr.delRate = req->start;
4669 log.u_bbr.cur_del_rate = req->end;
4670 log.u_bbr.flex1 = req->start_seq;
4671 log.u_bbr.flex2 = req->end_seq;
4672 log.u_bbr.flex3 = req->flags;
4673 log.u_bbr.flex4 = ((req->localtime >> 32) & 0x00000000ffffffff);
4674 log.u_bbr.flex5 = (req->localtime & 0x00000000ffffffff);
4675 log.u_bbr.flex7 = slot;
4676 log.u_bbr.bw_inuse = offset;
4677 /* nbytes = flex6 | epoch */
4678 log.u_bbr.flex6 = ((nbytes >> 32) & 0x00000000ffffffff);
4679 log.u_bbr.epoch = (nbytes & 0x00000000ffffffff);
4680 /* cspr = lt_epoch | pkts_out */
4681 log.u_bbr.lt_epoch = ((req->cspr >> 32) & 0x00000000ffffffff);
4682 log.u_bbr.pkts_out |= (req->cspr & 0x00000000ffffffff);
4683 log.u_bbr.applimited = tp->t_tcpreq_closed;
4684 log.u_bbr.applimited <<= 8;
4685 log.u_bbr.applimited |= tp->t_tcpreq_open;
4686 log.u_bbr.applimited <<= 8;
4687 log.u_bbr.applimited |= tp->t_tcpreq_req;
4688 log.u_bbr.timeStamp = tcp_get_usecs(&tv);
4689 TCP_LOG_EVENTP(tp, NULL,
4690 &tptosocket(tp)->so_rcv,
4691 &tptosocket(tp)->so_snd,
4692 TCP_LOG_REQ_T, 0,
4693 0, &log, false, &tv);
4694 }
4695 }
4696
4697 void
tcp_req_free_a_slot(struct tcpcb * tp,struct tcp_sendfile_track * ent)4698 tcp_req_free_a_slot(struct tcpcb *tp, struct tcp_sendfile_track *ent)
4699 {
4700 if (tp->t_tcpreq_req > 0)
4701 tp->t_tcpreq_req--;
4702 if (ent->flags & TCP_TRK_TRACK_FLG_OPEN) {
4703 if (tp->t_tcpreq_open > 0)
4704 tp->t_tcpreq_open--;
4705 } else {
4706 if (tp->t_tcpreq_closed > 0)
4707 tp->t_tcpreq_closed--;
4708 }
4709 ent->flags = TCP_TRK_TRACK_FLG_EMPTY;
4710 }
4711
4712 static void
tcp_req_check_for_stale_entries(struct tcpcb * tp,uint64_t ts,int rm_oldest)4713 tcp_req_check_for_stale_entries(struct tcpcb *tp, uint64_t ts, int rm_oldest)
4714 {
4715 struct tcp_sendfile_track *ent;
4716 uint64_t time_delta, oldest_delta;
4717 int i, oldest, oldest_set = 0, cnt_rm = 0;
4718
4719 for (i = 0; i < MAX_TCP_TRK_REQ; i++) {
4720 ent = &tp->t_tcpreq_info[i];
4721 if (ent->flags != TCP_TRK_TRACK_FLG_USED) {
4722 /*
4723 * We only care about closed end ranges
4724 * that are allocated and have no sendfile
4725 * ever touching them. They would be in
4726 * state USED.
4727 */
4728 continue;
4729 }
4730 if (ts >= ent->localtime)
4731 time_delta = ts - ent->localtime;
4732 else
4733 time_delta = 0;
4734 if (time_delta &&
4735 ((oldest_delta < time_delta) || (oldest_set == 0))) {
4736 oldest_set = 1;
4737 oldest = i;
4738 oldest_delta = time_delta;
4739 }
4740 if (tcp_stale_entry_time && (time_delta >= tcp_stale_entry_time)) {
4741 /*
4742 * No sendfile in a our time-limit
4743 * time to purge it.
4744 */
4745 cnt_rm++;
4746 tcp_req_log_req_info(tp, &tp->t_tcpreq_info[i], i, TCP_TRK_REQ_LOG_STALE,
4747 time_delta, 0);
4748 tcp_req_free_a_slot(tp, ent);
4749 }
4750 }
4751 if ((cnt_rm == 0) && rm_oldest && oldest_set) {
4752 ent = &tp->t_tcpreq_info[oldest];
4753 tcp_req_log_req_info(tp, &tp->t_tcpreq_info[i], i, TCP_TRK_REQ_LOG_STALE,
4754 oldest_delta, 1);
4755 tcp_req_free_a_slot(tp, ent);
4756 }
4757 }
4758
4759 int
tcp_req_check_for_comp(struct tcpcb * tp,tcp_seq ack_point)4760 tcp_req_check_for_comp(struct tcpcb *tp, tcp_seq ack_point)
4761 {
4762 int i, ret = 0;
4763 struct tcp_sendfile_track *ent;
4764
4765 /* Clean up any old closed end requests that are now completed */
4766 if (tp->t_tcpreq_req == 0)
4767 return (0);
4768 if (tp->t_tcpreq_closed == 0)
4769 return (0);
4770 for (i = 0; i < MAX_TCP_TRK_REQ; i++) {
4771 ent = &tp->t_tcpreq_info[i];
4772 /* Skip empty ones */
4773 if (ent->flags == TCP_TRK_TRACK_FLG_EMPTY)
4774 continue;
4775 /* Skip open ones */
4776 if (ent->flags & TCP_TRK_TRACK_FLG_OPEN)
4777 continue;
4778 if (SEQ_GEQ(ack_point, ent->end_seq)) {
4779 /* We are past it -- free it */
4780 tcp_req_log_req_info(tp, ent,
4781 i, TCP_TRK_REQ_LOG_FREED, 0, 0);
4782 tcp_req_free_a_slot(tp, ent);
4783 ret++;
4784 }
4785 }
4786 return (ret);
4787 }
4788
4789 int
tcp_req_is_entry_comp(struct tcpcb * tp,struct tcp_sendfile_track * ent,tcp_seq ack_point)4790 tcp_req_is_entry_comp(struct tcpcb *tp, struct tcp_sendfile_track *ent, tcp_seq ack_point)
4791 {
4792 if (tp->t_tcpreq_req == 0)
4793 return (-1);
4794 if (tp->t_tcpreq_closed == 0)
4795 return (-1);
4796 if (ent->flags == TCP_TRK_TRACK_FLG_EMPTY)
4797 return (-1);
4798 if (SEQ_GEQ(ack_point, ent->end_seq)) {
4799 return (1);
4800 }
4801 return (0);
4802 }
4803
4804 struct tcp_sendfile_track *
tcp_req_find_a_req_that_is_completed_by(struct tcpcb * tp,tcp_seq th_ack,int * ip)4805 tcp_req_find_a_req_that_is_completed_by(struct tcpcb *tp, tcp_seq th_ack, int *ip)
4806 {
4807 /*
4808 * Given an ack point (th_ack) walk through our entries and
4809 * return the first one found that th_ack goes past the
4810 * end_seq.
4811 */
4812 struct tcp_sendfile_track *ent;
4813 int i;
4814
4815 if (tp->t_tcpreq_req == 0) {
4816 /* none open */
4817 return (NULL);
4818 }
4819 for (i = 0; i < MAX_TCP_TRK_REQ; i++) {
4820 ent = &tp->t_tcpreq_info[i];
4821 if (ent->flags == TCP_TRK_TRACK_FLG_EMPTY)
4822 continue;
4823 if ((ent->flags & TCP_TRK_TRACK_FLG_OPEN) == 0) {
4824 if (SEQ_GEQ(th_ack, ent->end_seq)) {
4825 *ip = i;
4826 return (ent);
4827 }
4828 }
4829 }
4830 return (NULL);
4831 }
4832
4833 struct tcp_sendfile_track *
tcp_req_find_req_for_seq(struct tcpcb * tp,tcp_seq seq)4834 tcp_req_find_req_for_seq(struct tcpcb *tp, tcp_seq seq)
4835 {
4836 struct tcp_sendfile_track *ent;
4837 int i;
4838
4839 if (tp->t_tcpreq_req == 0) {
4840 /* none open */
4841 return (NULL);
4842 }
4843 for (i = 0; i < MAX_TCP_TRK_REQ; i++) {
4844 ent = &tp->t_tcpreq_info[i];
4845 tcp_req_log_req_info(tp, ent, i, TCP_TRK_REQ_LOG_SEARCH,
4846 (uint64_t)seq, 0);
4847 if (ent->flags == TCP_TRK_TRACK_FLG_EMPTY) {
4848 continue;
4849 }
4850 if (ent->flags & TCP_TRK_TRACK_FLG_OPEN) {
4851 /*
4852 * An open end request only needs to
4853 * match the beginning seq or be
4854 * all we have (once we keep going on
4855 * a open end request we may have a seq
4856 * wrap).
4857 */
4858 if ((SEQ_GEQ(seq, ent->start_seq)) ||
4859 (tp->t_tcpreq_closed == 0))
4860 return (ent);
4861 } else {
4862 /*
4863 * For this one we need to
4864 * be a bit more careful if its
4865 * completed at least.
4866 */
4867 if ((SEQ_GEQ(seq, ent->start_seq)) &&
4868 (SEQ_LT(seq, ent->end_seq))) {
4869 return (ent);
4870 }
4871 }
4872 }
4873 return (NULL);
4874 }
4875
4876 /* Should this be in its own file tcp_req.c ? */
4877 struct tcp_sendfile_track *
tcp_req_alloc_req_full(struct tcpcb * tp,struct tcp_snd_req * req,uint64_t ts,int rec_dups)4878 tcp_req_alloc_req_full(struct tcpcb *tp, struct tcp_snd_req *req, uint64_t ts, int rec_dups)
4879 {
4880 struct tcp_sendfile_track *fil;
4881 int i, allocated;
4882
4883 /* In case the stack does not check for completions do so now */
4884 tcp_req_check_for_comp(tp, tp->snd_una);
4885 /* Check for stale entries */
4886 if (tp->t_tcpreq_req)
4887 tcp_req_check_for_stale_entries(tp, ts,
4888 (tp->t_tcpreq_req >= MAX_TCP_TRK_REQ));
4889 /* Check to see if this is a duplicate of one not started */
4890 if (tp->t_tcpreq_req) {
4891 for (i = 0, allocated = 0; i < MAX_TCP_TRK_REQ; i++) {
4892 fil = &tp->t_tcpreq_info[i];
4893 if ((fil->flags & TCP_TRK_TRACK_FLG_USED) == 0)
4894 continue;
4895 if ((fil->timestamp == req->timestamp) &&
4896 (fil->start == req->start) &&
4897 ((fil->flags & TCP_TRK_TRACK_FLG_OPEN) ||
4898 (fil->end == req->end))) {
4899 /*
4900 * We already have this request
4901 * and it has not been started with sendfile.
4902 * This probably means the user was returned
4903 * a 4xx of some sort and its going to age
4904 * out, lets not duplicate it.
4905 */
4906 return (fil);
4907 }
4908 }
4909 }
4910 /* Ok if there is no room at the inn we are in trouble */
4911 if (tp->t_tcpreq_req >= MAX_TCP_TRK_REQ) {
4912 tcp_trace_point(tp, TCP_TP_REQ_LOG_FAIL);
4913 for (i = 0; i < MAX_TCP_TRK_REQ; i++) {
4914 tcp_req_log_req_info(tp, &tp->t_tcpreq_info[i],
4915 i, TCP_TRK_REQ_LOG_ALLOCFAIL, 0, 0);
4916 }
4917 return (NULL);
4918 }
4919 for (i = 0, allocated = 0; i < MAX_TCP_TRK_REQ; i++) {
4920 fil = &tp->t_tcpreq_info[i];
4921 if (fil->flags == TCP_TRK_TRACK_FLG_EMPTY) {
4922 allocated = 1;
4923 fil->flags = TCP_TRK_TRACK_FLG_USED;
4924 fil->timestamp = req->timestamp;
4925 fil->playout_ms = req->playout_ms;
4926 fil->localtime = ts;
4927 fil->start = req->start;
4928 if (req->flags & TCP_LOG_HTTPD_RANGE_END) {
4929 fil->end = req->end;
4930 } else {
4931 fil->end = 0;
4932 fil->flags |= TCP_TRK_TRACK_FLG_OPEN;
4933 }
4934 /*
4935 * We can set the min boundaries to the TCP Sequence space,
4936 * but it might be found to be further up when sendfile
4937 * actually runs on this range (if it ever does).
4938 */
4939 fil->sbcc_at_s = tptosocket(tp)->so_snd.sb_ccc;
4940 fil->start_seq = tp->snd_una +
4941 tptosocket(tp)->so_snd.sb_ccc;
4942 if (req->flags & TCP_LOG_HTTPD_RANGE_END)
4943 fil->end_seq = (fil->start_seq + ((uint32_t)(fil->end - fil->start)));
4944 else
4945 fil->end_seq = 0;
4946 if (tptosocket(tp)->so_snd.sb_tls_info) {
4947 /*
4948 * This session is doing TLS. Take a swag guess
4949 * at the overhead.
4950 */
4951 fil->end_seq += tcp_estimate_tls_overhead(
4952 tptosocket(tp), (fil->end - fil->start));
4953 }
4954 tp->t_tcpreq_req++;
4955 if (fil->flags & TCP_TRK_TRACK_FLG_OPEN)
4956 tp->t_tcpreq_open++;
4957 else
4958 tp->t_tcpreq_closed++;
4959 tcp_req_log_req_info(tp, fil, i,
4960 TCP_TRK_REQ_LOG_NEW, 0, 0);
4961 break;
4962 } else
4963 fil = NULL;
4964 }
4965 return (fil);
4966 }
4967
4968 void
tcp_req_alloc_req(struct tcpcb * tp,union tcp_log_userdata * user,uint64_t ts)4969 tcp_req_alloc_req(struct tcpcb *tp, union tcp_log_userdata *user, uint64_t ts)
4970 {
4971 (void)tcp_req_alloc_req_full(tp, &user->tcp_req, ts, 1);
4972 }
4973 #endif
4974
4975 void
tcp_log_socket_option(struct tcpcb * tp,uint32_t option_num,uint32_t option_val,int err)4976 tcp_log_socket_option(struct tcpcb *tp, uint32_t option_num, uint32_t option_val, int err)
4977 {
4978 if (tcp_bblogging_on(tp)) {
4979 struct tcp_log_buffer *l;
4980
4981 l = tcp_log_event(tp, NULL,
4982 &tptosocket(tp)->so_rcv,
4983 &tptosocket(tp)->so_snd,
4984 TCP_LOG_SOCKET_OPT,
4985 err, 0, NULL, 1,
4986 NULL, NULL, 0, NULL);
4987 if (l) {
4988 l->tlb_flex1 = option_num;
4989 l->tlb_flex2 = option_val;
4990 }
4991 }
4992 }
4993
4994 uint32_t
tcp_get_srtt(struct tcpcb * tp,int granularity)4995 tcp_get_srtt(struct tcpcb *tp, int granularity)
4996 {
4997 uint32_t srtt;
4998
4999 KASSERT(granularity == TCP_TMR_GRANULARITY_USEC ||
5000 granularity == TCP_TMR_GRANULARITY_TICKS,
5001 ("%s: called with unexpected granularity %d", __func__,
5002 granularity));
5003
5004 srtt = tp->t_srtt;
5005
5006 /*
5007 * We only support two granularities. If the stored granularity
5008 * does not match the granularity requested by the caller,
5009 * convert the stored value to the requested unit of granularity.
5010 */
5011 if (tp->t_tmr_granularity != granularity) {
5012 if (granularity == TCP_TMR_GRANULARITY_USEC)
5013 srtt = TICKS_2_USEC(srtt);
5014 else
5015 srtt = USEC_2_TICKS(srtt);
5016 }
5017
5018 /*
5019 * If the srtt is stored with ticks granularity, we need to
5020 * unshift to get the actual value. We do this after the
5021 * conversion above (if one was necessary) in order to maximize
5022 * precision.
5023 */
5024 if (tp->t_tmr_granularity == TCP_TMR_GRANULARITY_TICKS)
5025 srtt = srtt >> TCP_RTT_SHIFT;
5026
5027 return (srtt);
5028 }
5029
5030 void
tcp_account_for_send(struct tcpcb * tp,uint32_t len,uint8_t is_rxt,uint8_t is_tlp,bool hw_tls)5031 tcp_account_for_send(struct tcpcb *tp, uint32_t len, uint8_t is_rxt,
5032 uint8_t is_tlp, bool hw_tls)
5033 {
5034
5035 if (is_tlp) {
5036 tp->t_sndtlppack++;
5037 tp->t_sndtlpbyte += len;
5038 }
5039 /* To get total bytes sent you must add t_snd_rxt_bytes to t_sndbytes */
5040 if (is_rxt)
5041 tp->t_snd_rxt_bytes += len;
5042 else
5043 tp->t_sndbytes += len;
5044
5045 #ifdef KERN_TLS
5046 if (hw_tls && is_rxt && len != 0) {
5047 uint64_t rexmit_percent;
5048
5049 rexmit_percent = (1000ULL * tp->t_snd_rxt_bytes) /
5050 (10ULL * (tp->t_snd_rxt_bytes + tp->t_sndbytes));
5051 if (rexmit_percent > ktls_ifnet_max_rexmit_pct)
5052 ktls_disable_ifnet(tp);
5053 }
5054 #endif
5055 }
5056