1 /*-
2 * Copyright (c) 2001 McAfee, Inc.
3 * Copyright (c) 2006 Andre Oppermann, Internet Business Solutions AG
4 * All rights reserved.
5 *
6 * This software was developed for the FreeBSD Project by Jonathan Lemon
7 * and McAfee Research, the Security Research Division of McAfee, Inc. under
8 * DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the
9 * DARPA CHATS research program.
10 *
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
13 * are met:
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 */
32
33 #include <sys/cdefs.h>
34 __FBSDID("$FreeBSD$");
35
36 #include "opt_inet.h"
37 #include "opt_inet6.h"
38 #include "opt_ipsec.h"
39 #include "opt_mac.h"
40
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/kernel.h>
44 #include <sys/sysctl.h>
45 #include <sys/limits.h>
46 #include <sys/lock.h>
47 #include <sys/mutex.h>
48 #include <sys/malloc.h>
49 #include <sys/mbuf.h>
50 #include <sys/md5.h>
51 #include <sys/proc.h> /* for proc0 declaration */
52 #include <sys/random.h>
53 #include <sys/socket.h>
54 #include <sys/socketvar.h>
55 #include <sys/syslog.h>
56
57 #include <vm/uma.h>
58
59 #include <net/if.h>
60 #include <net/route.h>
61
62 #include <netinet/in.h>
63 #include <netinet/in_systm.h>
64 #include <netinet/ip.h>
65 #include <netinet/in_var.h>
66 #include <netinet/in_pcb.h>
67 #include <netinet/ip_var.h>
68 #include <netinet/ip_options.h>
69 #ifdef INET6
70 #include <netinet/ip6.h>
71 #include <netinet/icmp6.h>
72 #include <netinet6/nd6.h>
73 #include <netinet6/ip6_var.h>
74 #include <netinet6/in6_pcb.h>
75 #endif
76 #include <netinet/tcp.h>
77 #include <netinet/tcp_fsm.h>
78 #include <netinet/tcp_seq.h>
79 #include <netinet/tcp_timer.h>
80 #include <netinet/tcp_var.h>
81 #include <netinet/tcp_syncache.h>
82 #ifdef INET6
83 #include <netinet6/tcp6_var.h>
84 #endif
85
86 #ifdef IPSEC
87 #include <netipsec/ipsec.h>
88 #ifdef INET6
89 #include <netipsec/ipsec6.h>
90 #endif
91 #include <netipsec/key.h>
92 #endif /*IPSEC*/
93
94 #include <machine/in_cksum.h>
95
96 #include <security/mac/mac_framework.h>
97
98 static int tcp_syncookies = 1;
99 SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW,
100 &tcp_syncookies, 0,
101 "Use TCP SYN cookies if the syncache overflows");
102
103 static int tcp_syncookiesonly = 0;
104 SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies_only, CTLFLAG_RW,
105 &tcp_syncookiesonly, 0,
106 "Use only TCP SYN cookies");
107
108 #define SYNCOOKIE_SECRET_SIZE 8 /* dwords */
109 #define SYNCOOKIE_LIFETIME 16 /* seconds */
110
111 struct syncache {
112 TAILQ_ENTRY(syncache) sc_hash;
113 struct in_conninfo sc_inc; /* addresses */
114 int sc_rxttime; /* retransmit time */
115 u_int16_t sc_rxmits; /* retransmit counter */
116
117 u_int32_t sc_tsreflect; /* timestamp to reflect */
118 u_int32_t sc_ts; /* our timestamp to send */
119 u_int32_t sc_tsoff; /* ts offset w/ syncookies */
120 u_int32_t sc_flowlabel; /* IPv6 flowlabel */
121 tcp_seq sc_irs; /* seq from peer */
122 tcp_seq sc_iss; /* our ISS */
123 struct mbuf *sc_ipopts; /* source route */
124
125 u_int16_t sc_peer_mss; /* peer's MSS */
126 u_int16_t sc_wnd; /* advertised window */
127 u_int8_t sc_ip_ttl; /* IPv4 TTL */
128 u_int8_t sc_ip_tos; /* IPv4 TOS */
129 u_int8_t sc_requested_s_scale:4,
130 sc_requested_r_scale:4;
131 u_int8_t sc_flags;
132 #define SCF_NOOPT 0x01 /* no TCP options */
133 #define SCF_WINSCALE 0x02 /* negotiated window scaling */
134 #define SCF_TIMESTAMP 0x04 /* negotiated timestamps */
135 /* MSS is implicit */
136 #define SCF_UNREACH 0x10 /* icmp unreachable received */
137 #define SCF_SIGNATURE 0x20 /* send MD5 digests */
138 #define SCF_SACK 0x80 /* send SACK option */
139 #ifndef DISABLE_TCP_OFFLOAD
140 void *sc_pspare[2]; /* toepcb / toe_usrreqs */
141 #endif
142 #ifdef MAC
143 struct label *sc_label; /* MAC label reference */
144 #endif
145 };
146
147 struct syncache_head {
148 struct mtx sch_mtx;
149 TAILQ_HEAD(sch_head, syncache) sch_bucket;
150 struct callout sch_timer;
151 int sch_nextc;
152 u_int sch_length;
153 u_int sch_oddeven;
154 u_int32_t sch_secbits_odd[SYNCOOKIE_SECRET_SIZE];
155 u_int32_t sch_secbits_even[SYNCOOKIE_SECRET_SIZE];
156 u_int sch_reseed; /* time_uptime, seconds */
157 };
158
159 static void syncache_drop(struct syncache *, struct syncache_head *);
160 static void syncache_free(struct syncache *);
161 static void syncache_insert(struct syncache *, struct syncache_head *);
162 struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **);
163 static int syncache_respond(struct syncache *);
164 static struct socket *syncache_socket(struct syncache *, struct socket *,
165 struct mbuf *m);
166 static void syncache_timeout(struct syncache *sc, struct syncache_head *sch,
167 int docallout);
168 static void syncache_timer(void *);
169 static void syncookie_generate(struct syncache_head *, struct syncache *,
170 u_int32_t *);
171 static struct syncache
172 *syncookie_lookup(struct in_conninfo *, struct syncache_head *,
173 struct syncache *, struct tcpopt *, struct tcphdr *,
174 struct socket *);
175
176 /*
177 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies.
178 * 3 retransmits corresponds to a timeout of 3 * (1 + 2 + 4 + 8) == 45 seconds,
179 * the odds are that the user has given up attempting to connect by then.
180 */
181 #define SYNCACHE_MAXREXMTS 3
182
183 /* Arbitrary values */
184 #define TCP_SYNCACHE_HASHSIZE 512
185 #define TCP_SYNCACHE_BUCKETLIMIT 30
186
187 struct tcp_syncache {
188 struct syncache_head *hashbase;
189 uma_zone_t zone;
190 u_int hashsize;
191 u_int hashmask;
192 u_int bucket_limit;
193 u_int cache_count; /* XXX: unprotected */
194 u_int cache_limit;
195 u_int rexmt_limit;
196 u_int hash_secret;
197 };
198 static struct tcp_syncache tcp_syncache;
199
200 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache");
201
202 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RDTUN,
203 &tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache");
204
205 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RDTUN,
206 &tcp_syncache.cache_limit, 0, "Overall entry limit for syncache");
207
208 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD,
209 &tcp_syncache.cache_count, 0, "Current number of entries in syncache");
210
211 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RDTUN,
212 &tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable");
213
214 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW,
215 &tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions");
216
217 int tcp_sc_rst_sock_fail = 1;
218 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rst_on_sock_fail, CTLFLAG_RW,
219 &tcp_sc_rst_sock_fail, 0, "Send reset on socket allocation failure");
220
221 static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache");
222
223 #define SYNCACHE_HASH(inc, mask) \
224 ((tcp_syncache.hash_secret ^ \
225 (inc)->inc_faddr.s_addr ^ \
226 ((inc)->inc_faddr.s_addr >> 16) ^ \
227 (inc)->inc_fport ^ (inc)->inc_lport) & mask)
228
229 #define SYNCACHE_HASH6(inc, mask) \
230 ((tcp_syncache.hash_secret ^ \
231 (inc)->inc6_faddr.s6_addr32[0] ^ \
232 (inc)->inc6_faddr.s6_addr32[3] ^ \
233 (inc)->inc_fport ^ (inc)->inc_lport) & mask)
234
235 #define ENDPTS_EQ(a, b) ( \
236 (a)->ie_fport == (b)->ie_fport && \
237 (a)->ie_lport == (b)->ie_lport && \
238 (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr && \
239 (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr \
240 )
241
242 #define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0)
243
244 #define SCH_LOCK(sch) mtx_lock(&(sch)->sch_mtx)
245 #define SCH_UNLOCK(sch) mtx_unlock(&(sch)->sch_mtx)
246 #define SCH_LOCK_ASSERT(sch) mtx_assert(&(sch)->sch_mtx, MA_OWNED)
247
248 /*
249 * Requires the syncache entry to be already removed from the bucket list.
250 */
251 static void
252 syncache_free(struct syncache *sc)
253 {
254 if (sc->sc_ipopts)
255 (void) m_free(sc->sc_ipopts);
256 #ifdef MAC
257 mac_destroy_syncache(&sc->sc_label);
258 #endif
259
260 uma_zfree(tcp_syncache.zone, sc);
261 }
262
263 void
264 syncache_init(void)
265 {
266 int i;
267
268 tcp_syncache.cache_count = 0;
269 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
270 tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT;
271 tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS;
272 tcp_syncache.hash_secret = arc4random();
273
274 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize",
275 &tcp_syncache.hashsize);
276 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit",
277 &tcp_syncache.bucket_limit);
278 if (!powerof2(tcp_syncache.hashsize) || tcp_syncache.hashsize == 0) {
279 printf("WARNING: syncache hash size is not a power of 2.\n");
280 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
281 }
282 tcp_syncache.hashmask = tcp_syncache.hashsize - 1;
283
284 /* Set limits. */
285 tcp_syncache.cache_limit =
286 tcp_syncache.hashsize * tcp_syncache.bucket_limit;
287 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit",
288 &tcp_syncache.cache_limit);
289
290 /* Allocate the hash table. */
291 MALLOC(tcp_syncache.hashbase, struct syncache_head *,
292 tcp_syncache.hashsize * sizeof(struct syncache_head),
293 M_SYNCACHE, M_WAITOK | M_ZERO);
294
295 /* Initialize the hash buckets. */
296 for (i = 0; i < tcp_syncache.hashsize; i++) {
297 TAILQ_INIT(&tcp_syncache.hashbase[i].sch_bucket);
298 mtx_init(&tcp_syncache.hashbase[i].sch_mtx, "tcp_sc_head",
299 NULL, MTX_DEF);
300 callout_init_mtx(&tcp_syncache.hashbase[i].sch_timer,
301 &tcp_syncache.hashbase[i].sch_mtx, 0);
302 tcp_syncache.hashbase[i].sch_length = 0;
303 }
304
305 /* Create the syncache entry zone. */
306 tcp_syncache.zone = uma_zcreate("syncache", sizeof(struct syncache),
307 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
308 uma_zone_set_max(tcp_syncache.zone, tcp_syncache.cache_limit);
309 }
310
311 /*
312 * Inserts a syncache entry into the specified bucket row.
313 * Locks and unlocks the syncache_head autonomously.
314 */
315 static void
316 syncache_insert(struct syncache *sc, struct syncache_head *sch)
317 {
318 struct syncache *sc2;
319
320 SCH_LOCK(sch);
321
322 /*
323 * Make sure that we don't overflow the per-bucket limit.
324 * If the bucket is full, toss the oldest element.
325 */
326 if (sch->sch_length >= tcp_syncache.bucket_limit) {
327 KASSERT(!TAILQ_EMPTY(&sch->sch_bucket),
328 ("sch->sch_length incorrect"));
329 sc2 = TAILQ_LAST(&sch->sch_bucket, sch_head);
330 syncache_drop(sc2, sch);
331 tcpstat.tcps_sc_bucketoverflow++;
332 }
333
334 /* Put it into the bucket. */
335 TAILQ_INSERT_HEAD(&sch->sch_bucket, sc, sc_hash);
336 sch->sch_length++;
337
338 /* Reinitialize the bucket row's timer. */
339 if (sch->sch_length == 1)
340 sch->sch_nextc = ticks + INT_MAX;
341 syncache_timeout(sc, sch, 1);
342
343 SCH_UNLOCK(sch);
344
345 tcp_syncache.cache_count++;
346 tcpstat.tcps_sc_added++;
347 }
348
349 /*
350 * Remove and free entry from syncache bucket row.
351 * Expects locked syncache head.
352 */
353 static void
354 syncache_drop(struct syncache *sc, struct syncache_head *sch)
355 {
356
357 SCH_LOCK_ASSERT(sch);
358
359 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
360 sch->sch_length--;
361
362 syncache_free(sc);
363 tcp_syncache.cache_count--;
364 }
365
366 /*
367 * Engage/reengage time on bucket row.
368 */
369 static void
370 syncache_timeout(struct syncache *sc, struct syncache_head *sch, int docallout)
371 {
372 sc->sc_rxttime = ticks +
373 TCPTV_RTOBASE * (tcp_backoff[sc->sc_rxmits]);
374 sc->sc_rxmits++;
375 if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc)) {
376 sch->sch_nextc = sc->sc_rxttime;
377 if (docallout)
378 callout_reset(&sch->sch_timer, sch->sch_nextc - ticks,
379 syncache_timer, (void *)sch);
380 }
381 }
382
383 /*
384 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
385 * If we have retransmitted an entry the maximum number of times, expire it.
386 * One separate timer for each bucket row.
387 */
388 static void
389 syncache_timer(void *xsch)
390 {
391 struct syncache_head *sch = (struct syncache_head *)xsch;
392 struct syncache *sc, *nsc;
393 int tick = ticks;
394 char *s;
395
396 /* NB: syncache_head has already been locked by the callout. */
397 SCH_LOCK_ASSERT(sch);
398
399 /*
400 * In the following cycle we may remove some entries and/or
401 * advance some timeouts, so re-initialize the bucket timer.
402 */
403 sch->sch_nextc = tick + INT_MAX;
404
405 TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc) {
406 /*
407 * We do not check if the listen socket still exists
408 * and accept the case where the listen socket may be
409 * gone by the time we resend the SYN/ACK. We do
410 * not expect this to happens often. If it does,
411 * then the RST will be sent by the time the remote
412 * host does the SYN/ACK->ACK.
413 */
414 if (TSTMP_GT(sc->sc_rxttime, tick)) {
415 if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc))
416 sch->sch_nextc = sc->sc_rxttime;
417 continue;
418 }
419
420 if (sc->sc_rxmits > tcp_syncache.rexmt_limit) {
421 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
422 log(LOG_DEBUG, "%s; %s: Retransmits exhausted, "
423 "giving up and removing syncache entry\n",
424 s, __func__);
425 free(s, M_TCPLOG);
426 }
427 syncache_drop(sc, sch);
428 tcpstat.tcps_sc_stale++;
429 continue;
430 }
431 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
432 log(LOG_DEBUG, "%s; %s: Response timeout, "
433 "retransmitting (%u) SYN|ACK\n",
434 s, __func__, sc->sc_rxmits);
435 free(s, M_TCPLOG);
436 }
437
438 (void) syncache_respond(sc);
439 tcpstat.tcps_sc_retransmitted++;
440 syncache_timeout(sc, sch, 0);
441 }
442 if (!TAILQ_EMPTY(&(sch)->sch_bucket))
443 callout_reset(&(sch)->sch_timer, (sch)->sch_nextc - tick,
444 syncache_timer, (void *)(sch));
445 }
446
447 /*
448 * Find an entry in the syncache.
449 * Returns always with locked syncache_head plus a matching entry or NULL.
450 */
451 struct syncache *
452 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp)
453 {
454 struct syncache *sc;
455 struct syncache_head *sch;
456
457 #ifdef INET6
458 if (inc->inc_isipv6) {
459 sch = &tcp_syncache.hashbase[
460 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)];
461 *schp = sch;
462
463 SCH_LOCK(sch);
464
465 /* Circle through bucket row to find matching entry. */
466 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
467 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
468 return (sc);
469 }
470 } else
471 #endif
472 {
473 sch = &tcp_syncache.hashbase[
474 SYNCACHE_HASH(inc, tcp_syncache.hashmask)];
475 *schp = sch;
476
477 SCH_LOCK(sch);
478
479 /* Circle through bucket row to find matching entry. */
480 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
481 #ifdef INET6
482 if (sc->sc_inc.inc_isipv6)
483 continue;
484 #endif
485 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
486 return (sc);
487 }
488 }
489 SCH_LOCK_ASSERT(*schp);
490 return (NULL); /* always returns with locked sch */
491 }
492
493 /*
494 * This function is called when we get a RST for a
495 * non-existent connection, so that we can see if the
496 * connection is in the syn cache. If it is, zap it.
497 */
498 void
499 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th)
500 {
501 struct syncache *sc;
502 struct syncache_head *sch;
503 char *s = NULL;
504
505 sc = syncache_lookup(inc, &sch); /* returns locked sch */
506 SCH_LOCK_ASSERT(sch);
507
508 /*
509 * Any RST to our SYN|ACK must not carry ACK, SYN or FIN flags.
510 * See RFC 793 page 65, section SEGMENT ARRIVES.
511 */
512 if (th->th_flags & (TH_ACK|TH_SYN|TH_FIN)) {
513 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
514 log(LOG_DEBUG, "%s; %s: Spurious RST with ACK, SYN or "
515 "FIN flag set, segment ignored\n", s, __func__);
516 tcpstat.tcps_badrst++;
517 goto done;
518 }
519
520 /*
521 * No corresponding connection was found in syncache.
522 * If syncookies are enabled and possibly exclusively
523 * used, or we are under memory pressure, a valid RST
524 * may not find a syncache entry. In that case we're
525 * done and no SYN|ACK retransmissions will happen.
526 * Otherwise the the RST was misdirected or spoofed.
527 */
528 if (sc == NULL) {
529 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
530 log(LOG_DEBUG, "%s; %s: Spurious RST without matching "
531 "syncache entry (possibly syncookie only), "
532 "segment ignored\n", s, __func__);
533 tcpstat.tcps_badrst++;
534 goto done;
535 }
536
537 /*
538 * If the RST bit is set, check the sequence number to see
539 * if this is a valid reset segment.
540 * RFC 793 page 37:
541 * In all states except SYN-SENT, all reset (RST) segments
542 * are validated by checking their SEQ-fields. A reset is
543 * valid if its sequence number is in the window.
544 *
545 * The sequence number in the reset segment is normally an
546 * echo of our outgoing acknowlegement numbers, but some hosts
547 * send a reset with the sequence number at the rightmost edge
548 * of our receive window, and we have to handle this case.
549 */
550 if (SEQ_GEQ(th->th_seq, sc->sc_irs) &&
551 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) {
552 syncache_drop(sc, sch);
553 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
554 log(LOG_DEBUG, "%s; %s: Our SYN|ACK was rejected, "
555 "connection attempt aborted by remote endpoint\n",
556 s, __func__);
557 tcpstat.tcps_sc_reset++;
558 } else if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
559 log(LOG_DEBUG, "%s; %s: RST with invalid SEQ %u != IRS %u "
560 "(+WND %u), segment ignored\n",
561 s, __func__, th->th_seq, sc->sc_irs, sc->sc_wnd);
562 tcpstat.tcps_badrst++;
563 }
564
565 done:
566 if (s != NULL)
567 free(s, M_TCPLOG);
568 SCH_UNLOCK(sch);
569 }
570
571 void
572 syncache_badack(struct in_conninfo *inc)
573 {
574 struct syncache *sc;
575 struct syncache_head *sch;
576
577 sc = syncache_lookup(inc, &sch); /* returns locked sch */
578 SCH_LOCK_ASSERT(sch);
579 if (sc != NULL) {
580 syncache_drop(sc, sch);
581 tcpstat.tcps_sc_badack++;
582 }
583 SCH_UNLOCK(sch);
584 }
585
586 void
587 syncache_unreach(struct in_conninfo *inc, struct tcphdr *th)
588 {
589 struct syncache *sc;
590 struct syncache_head *sch;
591
592 sc = syncache_lookup(inc, &sch); /* returns locked sch */
593 SCH_LOCK_ASSERT(sch);
594 if (sc == NULL)
595 goto done;
596
597 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */
598 if (ntohl(th->th_seq) != sc->sc_iss)
599 goto done;
600
601 /*
602 * If we've rertransmitted 3 times and this is our second error,
603 * we remove the entry. Otherwise, we allow it to continue on.
604 * This prevents us from incorrectly nuking an entry during a
605 * spurious network outage.
606 *
607 * See tcp_notify().
608 */
609 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxmits < 3 + 1) {
610 sc->sc_flags |= SCF_UNREACH;
611 goto done;
612 }
613 syncache_drop(sc, sch);
614 tcpstat.tcps_sc_unreach++;
615 done:
616 SCH_UNLOCK(sch);
617 }
618
619 /*
620 * Build a new TCP socket structure from a syncache entry.
621 */
622 static struct socket *
623 syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m)
624 {
625 struct inpcb *inp = NULL;
626 struct socket *so;
627 struct tcpcb *tp;
628 char *s;
629
630 INP_INFO_WLOCK_ASSERT(&tcbinfo);
631
632 /*
633 * Ok, create the full blown connection, and set things up
634 * as they would have been set up if we had created the
635 * connection when the SYN arrived. If we can't create
636 * the connection, abort it.
637 */
638 so = sonewconn(lso, SS_ISCONNECTED);
639 if (so == NULL) {
640 /*
641 * Drop the connection; we will either send a RST or
642 * have the peer retransmit its SYN again after its
643 * RTO and try again.
644 */
645 tcpstat.tcps_listendrop++;
646 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
647 log(LOG_DEBUG, "%s; %s: Socket create failed "
648 "due to limits or memory shortage\n",
649 s, __func__);
650 free(s, M_TCPLOG);
651 }
652 goto abort2;
653 }
654 #ifdef MAC
655 SOCK_LOCK(so);
656 mac_set_socket_peer_from_mbuf(m, so);
657 SOCK_UNLOCK(so);
658 #endif
659
660 inp = sotoinpcb(so);
661 INP_LOCK(inp);
662
663 /* Insert new socket into PCB hash list. */
664 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6;
665 #ifdef INET6
666 if (sc->sc_inc.inc_isipv6) {
667 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
668 } else {
669 inp->inp_vflag &= ~INP_IPV6;
670 inp->inp_vflag |= INP_IPV4;
671 #endif
672 inp->inp_laddr = sc->sc_inc.inc_laddr;
673 #ifdef INET6
674 }
675 #endif
676 inp->inp_lport = sc->sc_inc.inc_lport;
677 if (in_pcbinshash(inp) != 0) {
678 /*
679 * Undo the assignments above if we failed to
680 * put the PCB on the hash lists.
681 */
682 #ifdef INET6
683 if (sc->sc_inc.inc_isipv6)
684 inp->in6p_laddr = in6addr_any;
685 else
686 #endif
687 inp->inp_laddr.s_addr = INADDR_ANY;
688 inp->inp_lport = 0;
689 goto abort;
690 }
691 #ifdef IPSEC
692 /* Copy old policy into new socket's. */
693 if (ipsec_copy_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp))
694 printf("syncache_socket: could not copy policy\n");
695 #endif
696 #ifdef INET6
697 if (sc->sc_inc.inc_isipv6) {
698 struct inpcb *oinp = sotoinpcb(lso);
699 struct in6_addr laddr6;
700 struct sockaddr_in6 sin6;
701 /*
702 * Inherit socket options from the listening socket.
703 * Note that in6p_inputopts are not (and should not be)
704 * copied, since it stores previously received options and is
705 * used to detect if each new option is different than the
706 * previous one and hence should be passed to a user.
707 * If we copied in6p_inputopts, a user would not be able to
708 * receive options just after calling the accept system call.
709 */
710 inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS;
711 if (oinp->in6p_outputopts)
712 inp->in6p_outputopts =
713 ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT);
714
715 sin6.sin6_family = AF_INET6;
716 sin6.sin6_len = sizeof(sin6);
717 sin6.sin6_addr = sc->sc_inc.inc6_faddr;
718 sin6.sin6_port = sc->sc_inc.inc_fport;
719 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0;
720 laddr6 = inp->in6p_laddr;
721 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
722 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
723 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6,
724 thread0.td_ucred)) {
725 inp->in6p_laddr = laddr6;
726 goto abort;
727 }
728 /* Override flowlabel from in6_pcbconnect. */
729 inp->in6p_flowinfo &= ~IPV6_FLOWLABEL_MASK;
730 inp->in6p_flowinfo |= sc->sc_flowlabel;
731 } else
732 #endif
733 {
734 struct in_addr laddr;
735 struct sockaddr_in sin;
736
737 inp->inp_options = ip_srcroute(m);
738 if (inp->inp_options == NULL) {
739 inp->inp_options = sc->sc_ipopts;
740 sc->sc_ipopts = NULL;
741 }
742
743 sin.sin_family = AF_INET;
744 sin.sin_len = sizeof(sin);
745 sin.sin_addr = sc->sc_inc.inc_faddr;
746 sin.sin_port = sc->sc_inc.inc_fport;
747 bzero((caddr_t)sin.sin_zero, sizeof(sin.sin_zero));
748 laddr = inp->inp_laddr;
749 if (inp->inp_laddr.s_addr == INADDR_ANY)
750 inp->inp_laddr = sc->sc_inc.inc_laddr;
751 if (in_pcbconnect(inp, (struct sockaddr *)&sin,
752 thread0.td_ucred)) {
753 inp->inp_laddr = laddr;
754 goto abort;
755 }
756 }
757 tp = intotcpcb(inp);
758 tp->t_state = TCPS_SYN_RECEIVED;
759 tp->iss = sc->sc_iss;
760 tp->irs = sc->sc_irs;
761 tcp_rcvseqinit(tp);
762 tcp_sendseqinit(tp);
763 tp->snd_wl1 = sc->sc_irs;
764 tp->snd_max = tp->iss + 1;
765 tp->snd_nxt = tp->iss + 1;
766 tp->rcv_up = sc->sc_irs + 1;
767 tp->rcv_wnd = sc->sc_wnd;
768 tp->rcv_adv += tp->rcv_wnd;
769 tp->last_ack_sent = tp->rcv_nxt;
770
771 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY);
772 if (sc->sc_flags & SCF_NOOPT)
773 tp->t_flags |= TF_NOOPT;
774 else {
775 if (sc->sc_flags & SCF_WINSCALE) {
776 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE;
777 tp->snd_scale = sc->sc_requested_s_scale;
778 tp->request_r_scale = sc->sc_requested_r_scale;
779 }
780 if (sc->sc_flags & SCF_TIMESTAMP) {
781 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP;
782 tp->ts_recent = sc->sc_tsreflect;
783 tp->ts_recent_age = ticks;
784 tp->ts_offset = sc->sc_tsoff;
785 }
786 #ifdef TCP_SIGNATURE
787 if (sc->sc_flags & SCF_SIGNATURE)
788 tp->t_flags |= TF_SIGNATURE;
789 #endif
790 if (sc->sc_flags & SCF_SACK)
791 tp->t_flags |= TF_SACK_PERMIT;
792 }
793
794 /*
795 * Set up MSS and get cached values from tcp_hostcache.
796 * This might overwrite some of the defaults we just set.
797 */
798 tcp_mss(tp, sc->sc_peer_mss);
799
800 /*
801 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment.
802 */
803 if (sc->sc_rxmits)
804 tp->snd_cwnd = tp->t_maxseg;
805 tcp_timer_activate(tp, TT_KEEP, tcp_keepinit);
806
807 INP_UNLOCK(inp);
808
809 tcpstat.tcps_accepts++;
810 return (so);
811
812 abort:
813 INP_UNLOCK(inp);
814 abort2:
815 if (so != NULL)
816 soabort(so);
817 return (NULL);
818 }
819
820 /*
821 * This function gets called when we receive an ACK for a
822 * socket in the LISTEN state. We look up the connection
823 * in the syncache, and if its there, we pull it out of
824 * the cache and turn it into a full-blown connection in
825 * the SYN-RECEIVED state.
826 */
827 int
828 syncache_expand(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
829 struct socket **lsop, struct mbuf *m)
830 {
831 struct syncache *sc;
832 struct syncache_head *sch;
833 struct syncache scs;
834 char *s;
835
836 /*
837 * Global TCP locks are held because we manipulate the PCB lists
838 * and create a new socket.
839 */
840 INP_INFO_WLOCK_ASSERT(&tcbinfo);
841 KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK,
842 ("%s: can handle only ACK", __func__));
843
844 sc = syncache_lookup(inc, &sch); /* returns locked sch */
845 SCH_LOCK_ASSERT(sch);
846 if (sc == NULL) {
847 /*
848 * There is no syncache entry, so see if this ACK is
849 * a returning syncookie. To do this, first:
850 * A. See if this socket has had a syncache entry dropped in
851 * the past. We don't want to accept a bogus syncookie
852 * if we've never received a SYN.
853 * B. check that the syncookie is valid. If it is, then
854 * cobble up a fake syncache entry, and return.
855 */
856 if (!tcp_syncookies) {
857 SCH_UNLOCK(sch);
858 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
859 log(LOG_DEBUG, "%s; %s: Spurious ACK, "
860 "segment rejected (syncookies disabled)\n",
861 s, __func__);
862 goto failed;
863 }
864 bzero(&scs, sizeof(scs));
865 sc = syncookie_lookup(inc, sch, &scs, to, th, *lsop);
866 SCH_UNLOCK(sch);
867 if (sc == NULL) {
868 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
869 log(LOG_DEBUG, "%s; %s: Segment failed "
870 "SYNCOOKIE authentication, segment rejected "
871 "(probably spoofed)\n", s, __func__);
872 goto failed;
873 }
874 } else {
875 /* Pull out the entry to unlock the bucket row. */
876 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
877 sch->sch_length--;
878 tcp_syncache.cache_count--;
879 SCH_UNLOCK(sch);
880 }
881
882 /*
883 * Segment validation:
884 * ACK must match our initial sequence number + 1 (the SYN|ACK).
885 */
886 if (th->th_ack != sc->sc_iss + 1) {
887 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
888 log(LOG_DEBUG, "%s; %s: ACK %u != ISS+1 %u, segment "
889 "rejected\n", s, __func__, th->th_ack, sc->sc_iss);
890 goto failed;
891 }
892 /*
893 * The SEQ must match the received initial receive sequence
894 * number + 1 (the SYN) because we didn't ACK any data that
895 * may have come with the SYN.
896 */
897 if (th->th_seq != sc->sc_irs + 1) {
898 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
899 log(LOG_DEBUG, "%s; %s: SEQ %u != IRS+1 %u, segment "
900 "rejected\n", s, __func__, th->th_seq, sc->sc_irs);
901 goto failed;
902 }
903 #if 0
904 /*
905 * If timestamps were present in the SYN and we accepted
906 * them in our SYN|ACK we require them to be present from
907 * now on. And vice versa.
908 *
909 * Unfortunately, during testing of 7.0 some users found
910 * network devices that violate this constraint, so it must
911 * be disabled.
912 */
913 if ((sc->sc_flags & SCF_TIMESTAMP) && !(to->to_flags & TOF_TS)) {
914 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
915 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
916 "segment rejected\n", s, __func__);
917 goto failed;
918 }
919 #endif
920 if (!(sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS)) {
921 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
922 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
923 "segment rejected\n", s, __func__);
924 goto failed;
925 }
926 /*
927 * If timestamps were negotiated the reflected timestamp
928 * must be equal to what we actually sent in the SYN|ACK.
929 */
930 if ((to->to_flags & TOF_TS) && to->to_tsecr != sc->sc_ts) {
931 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
932 log(LOG_DEBUG, "%s; %s: TSECR %u != TS %u, "
933 "segment rejected\n",
934 s, __func__, to->to_tsecr, sc->sc_ts);
935 goto failed;
936 }
937
938 *lsop = syncache_socket(sc, *lsop, m);
939
940 if (*lsop == NULL)
941 tcpstat.tcps_sc_aborted++;
942 else
943 tcpstat.tcps_sc_completed++;
944
945 if (sc != &scs)
946 syncache_free(sc);
947 return (1);
948 failed:
949 if (sc != NULL && sc != &scs)
950 syncache_free(sc);
951 if (s != NULL)
952 free(s, M_TCPLOG);
953 *lsop = NULL;
954 return (0);
955 }
956
957 /*
958 * Given a LISTEN socket and an inbound SYN request, add
959 * this to the syn cache, and send back a segment:
960 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
961 * to the source.
962 *
963 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
964 * Doing so would require that we hold onto the data and deliver it
965 * to the application. However, if we are the target of a SYN-flood
966 * DoS attack, an attacker could send data which would eventually
967 * consume all available buffer space if it were ACKed. By not ACKing
968 * the data, we avoid this DoS scenario.
969 */
970 void
971 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
972 struct inpcb *inp, struct socket **lsop, struct mbuf *m)
973 {
974 struct tcpcb *tp;
975 struct socket *so;
976 struct syncache *sc = NULL;
977 struct syncache_head *sch;
978 struct mbuf *ipopts = NULL;
979 u_int32_t flowtmp;
980 int win, sb_hiwat, ip_ttl, ip_tos, noopt;
981 char *s;
982 #ifdef INET6
983 int autoflowlabel = 0;
984 #endif
985 #ifdef MAC
986 struct label *maclabel;
987 #endif
988 struct syncache scs;
989
990 INP_INFO_WLOCK_ASSERT(&tcbinfo);
991 INP_LOCK_ASSERT(inp); /* listen socket */
992 KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN,
993 ("%s: unexpected tcp flags", __func__));
994
995 /*
996 * Combine all so/tp operations very early to drop the INP lock as
997 * soon as possible.
998 */
999 so = *lsop;
1000 tp = sototcpcb(so);
1001
1002 #ifdef INET6
1003 if (inc->inc_isipv6 &&
1004 (inp->in6p_flags & IN6P_AUTOFLOWLABEL))
1005 autoflowlabel = 1;
1006 #endif
1007 ip_ttl = inp->inp_ip_ttl;
1008 ip_tos = inp->inp_ip_tos;
1009 win = sbspace(&so->so_rcv);
1010 sb_hiwat = so->so_rcv.sb_hiwat;
1011 noopt = (tp->t_flags & TF_NOOPT);
1012
1013 so = NULL;
1014 tp = NULL;
1015
1016 #ifdef MAC
1017 if (mac_init_syncache(&maclabel) != 0) {
1018 INP_UNLOCK(inp);
1019 INP_INFO_WUNLOCK(&tcbinfo);
1020 goto done;
1021 } else
1022 mac_init_syncache_from_inpcb(maclabel, inp);
1023 #endif
1024 INP_UNLOCK(inp);
1025 INP_INFO_WUNLOCK(&tcbinfo);
1026
1027 /*
1028 * Remember the IP options, if any.
1029 */
1030 #ifdef INET6
1031 if (!inc->inc_isipv6)
1032 #endif
1033 ipopts = ip_srcroute(m);
1034
1035 /*
1036 * See if we already have an entry for this connection.
1037 * If we do, resend the SYN,ACK, and reset the retransmit timer.
1038 *
1039 * XXX: should the syncache be re-initialized with the contents
1040 * of the new SYN here (which may have different options?)
1041 *
1042 * XXX: We do not check the sequence number to see if this is a
1043 * real retransmit or a new connection attempt. The question is
1044 * how to handle such a case; either ignore it as spoofed, or
1045 * drop the current entry and create a new one?
1046 */
1047 sc = syncache_lookup(inc, &sch); /* returns locked entry */
1048 SCH_LOCK_ASSERT(sch);
1049 if (sc != NULL) {
1050 tcpstat.tcps_sc_dupsyn++;
1051 if (ipopts) {
1052 /*
1053 * If we were remembering a previous source route,
1054 * forget it and use the new one we've been given.
1055 */
1056 if (sc->sc_ipopts)
1057 (void) m_free(sc->sc_ipopts);
1058 sc->sc_ipopts = ipopts;
1059 }
1060 /*
1061 * Update timestamp if present.
1062 */
1063 if ((sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS))
1064 sc->sc_tsreflect = to->to_tsval;
1065 else
1066 sc->sc_flags &= ~SCF_TIMESTAMP;
1067 #ifdef MAC
1068 /*
1069 * Since we have already unconditionally allocated label
1070 * storage, free it up. The syncache entry will already
1071 * have an initialized label we can use.
1072 */
1073 mac_destroy_syncache(&maclabel);
1074 KASSERT(sc->sc_label != NULL,
1075 ("%s: label not initialized", __func__));
1076 #endif
1077 /* Retransmit SYN|ACK and reset retransmit count. */
1078 if ((s = tcp_log_addrs(&sc->sc_inc, th, NULL, NULL))) {
1079 log(LOG_DEBUG, "%s; %s: Received duplicate SYN, "
1080 "resetting timer and retransmitting SYN|ACK\n",
1081 s, __func__);
1082 free(s, M_TCPLOG);
1083 }
1084 if (syncache_respond(sc) == 0) {
1085 sc->sc_rxmits = 0;
1086 syncache_timeout(sc, sch, 1);
1087 tcpstat.tcps_sndacks++;
1088 tcpstat.tcps_sndtotal++;
1089 }
1090 SCH_UNLOCK(sch);
1091 goto done;
1092 }
1093
1094 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT | M_ZERO);
1095 if (sc == NULL) {
1096 /*
1097 * The zone allocator couldn't provide more entries.
1098 * Treat this as if the cache was full; drop the oldest
1099 * entry and insert the new one.
1100 */
1101 tcpstat.tcps_sc_zonefail++;
1102 if ((sc = TAILQ_LAST(&sch->sch_bucket, sch_head)) != NULL)
1103 syncache_drop(sc, sch);
1104 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT | M_ZERO);
1105 if (sc == NULL) {
1106 if (tcp_syncookies) {
1107 bzero(&scs, sizeof(scs));
1108 sc = &scs;
1109 } else {
1110 SCH_UNLOCK(sch);
1111 if (ipopts)
1112 (void) m_free(ipopts);
1113 goto done;
1114 }
1115 }
1116 }
1117
1118 /*
1119 * Fill in the syncache values.
1120 */
1121 #ifdef MAC
1122 sc->sc_label = maclabel;
1123 #endif
1124 sc->sc_ipopts = ipopts;
1125 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo));
1126 #ifdef INET6
1127 if (!inc->inc_isipv6)
1128 #endif
1129 {
1130 sc->sc_ip_tos = ip_tos;
1131 sc->sc_ip_ttl = ip_ttl;
1132 }
1133
1134 sc->sc_irs = th->th_seq;
1135 sc->sc_iss = arc4random();
1136 sc->sc_flags = 0;
1137 sc->sc_flowlabel = 0;
1138
1139 /*
1140 * Initial receive window: clip sbspace to [0 .. TCP_MAXWIN].
1141 * win was derived from socket earlier in the function.
1142 */
1143 win = imax(win, 0);
1144 win = imin(win, TCP_MAXWIN);
1145 sc->sc_wnd = win;
1146
1147 if (tcp_do_rfc1323) {
1148 /*
1149 * A timestamp received in a SYN makes
1150 * it ok to send timestamp requests and replies.
1151 */
1152 if (to->to_flags & TOF_TS) {
1153 sc->sc_tsreflect = to->to_tsval;
1154 sc->sc_ts = ticks;
1155 sc->sc_flags |= SCF_TIMESTAMP;
1156 }
1157 if (to->to_flags & TOF_SCALE) {
1158 int wscale = 0;
1159
1160 /*
1161 * Pick the smallest possible scaling factor that
1162 * will still allow us to scale up to sb_max, aka
1163 * kern.ipc.maxsockbuf.
1164 *
1165 * We do this because there are broken firewalls that
1166 * will corrupt the window scale option, leading to
1167 * the other endpoint believing that our advertised
1168 * window is unscaled. At scale factors larger than
1169 * 5 the unscaled window will drop below 1500 bytes,
1170 * leading to serious problems when traversing these
1171 * broken firewalls.
1172 *
1173 * With the default maxsockbuf of 256K, a scale factor
1174 * of 3 will be chosen by this algorithm. Those who
1175 * choose a larger maxsockbuf should watch out
1176 * for the compatiblity problems mentioned above.
1177 *
1178 * RFC1323: The Window field in a SYN (i.e., a <SYN>
1179 * or <SYN,ACK>) segment itself is never scaled.
1180 */
1181 while (wscale < TCP_MAX_WINSHIFT &&
1182 (TCP_MAXWIN << wscale) < sb_max)
1183 wscale++;
1184 sc->sc_requested_r_scale = wscale;
1185 sc->sc_requested_s_scale = to->to_wscale;
1186 sc->sc_flags |= SCF_WINSCALE;
1187 }
1188 }
1189 #ifdef TCP_SIGNATURE
1190 /*
1191 * If listening socket requested TCP digests, and received SYN
1192 * contains the option, flag this in the syncache so that
1193 * syncache_respond() will do the right thing with the SYN+ACK.
1194 * XXX: Currently we always record the option by default and will
1195 * attempt to use it in syncache_respond().
1196 */
1197 if (to->to_flags & TOF_SIGNATURE)
1198 sc->sc_flags |= SCF_SIGNATURE;
1199 #endif
1200 if (to->to_flags & TOF_SACKPERM)
1201 sc->sc_flags |= SCF_SACK;
1202 if (to->to_flags & TOF_MSS)
1203 sc->sc_peer_mss = to->to_mss; /* peer mss may be zero */
1204 if (noopt)
1205 sc->sc_flags |= SCF_NOOPT;
1206
1207 if (tcp_syncookies) {
1208 syncookie_generate(sch, sc, &flowtmp);
1209 #ifdef INET6
1210 if (autoflowlabel)
1211 sc->sc_flowlabel = flowtmp;
1212 #endif
1213 } else {
1214 #ifdef INET6
1215 if (autoflowlabel)
1216 sc->sc_flowlabel =
1217 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
1218 #endif
1219 }
1220 SCH_UNLOCK(sch);
1221
1222 /*
1223 * Do a standard 3-way handshake.
1224 */
1225 if (syncache_respond(sc) == 0) {
1226 if (tcp_syncookies && tcp_syncookiesonly && sc != &scs)
1227 syncache_free(sc);
1228 else if (sc != &scs)
1229 syncache_insert(sc, sch); /* locks and unlocks sch */
1230 tcpstat.tcps_sndacks++;
1231 tcpstat.tcps_sndtotal++;
1232 } else {
1233 if (sc != &scs)
1234 syncache_free(sc);
1235 tcpstat.tcps_sc_dropped++;
1236 }
1237
1238 done:
1239 #ifdef MAC
1240 if (sc == &scs)
1241 mac_destroy_syncache(&maclabel);
1242 #endif
1243 *lsop = NULL;
1244 m_freem(m);
1245 return;
1246 }
1247
1248 static int
1249 syncache_respond(struct syncache *sc)
1250 {
1251 struct ip *ip = NULL;
1252 struct mbuf *m;
1253 struct tcphdr *th;
1254 int optlen, error;
1255 u_int16_t hlen, tlen, mssopt;
1256 struct tcpopt to;
1257 #ifdef INET6
1258 struct ip6_hdr *ip6 = NULL;
1259 #endif
1260
1261 hlen =
1262 #ifdef INET6
1263 (sc->sc_inc.inc_isipv6) ? sizeof(struct ip6_hdr) :
1264 #endif
1265 sizeof(struct ip);
1266 tlen = hlen + sizeof(struct tcphdr);
1267
1268 /* Determine MSS we advertize to other end of connection. */
1269 mssopt = tcp_mssopt(&sc->sc_inc);
1270 if (sc->sc_peer_mss)
1271 mssopt = max( min(sc->sc_peer_mss, mssopt), tcp_minmss);
1272
1273 /* XXX: Assume that the entire packet will fit in a header mbuf. */
1274 KASSERT(max_linkhdr + tlen + TCP_MAXOLEN <= MHLEN,
1275 ("syncache: mbuf too small"));
1276
1277 /* Create the IP+TCP header from scratch. */
1278 m = m_gethdr(M_DONTWAIT, MT_DATA);
1279 if (m == NULL)
1280 return (ENOBUFS);
1281 #ifdef MAC
1282 mac_create_mbuf_from_syncache(sc->sc_label, m);
1283 #endif
1284 m->m_data += max_linkhdr;
1285 m->m_len = tlen;
1286 m->m_pkthdr.len = tlen;
1287 m->m_pkthdr.rcvif = NULL;
1288
1289 #ifdef INET6
1290 if (sc->sc_inc.inc_isipv6) {
1291 ip6 = mtod(m, struct ip6_hdr *);
1292 ip6->ip6_vfc = IPV6_VERSION;
1293 ip6->ip6_nxt = IPPROTO_TCP;
1294 ip6->ip6_src = sc->sc_inc.inc6_laddr;
1295 ip6->ip6_dst = sc->sc_inc.inc6_faddr;
1296 ip6->ip6_plen = htons(tlen - hlen);
1297 /* ip6_hlim is set after checksum */
1298 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
1299 ip6->ip6_flow |= sc->sc_flowlabel;
1300
1301 th = (struct tcphdr *)(ip6 + 1);
1302 } else
1303 #endif
1304 {
1305 ip = mtod(m, struct ip *);
1306 ip->ip_v = IPVERSION;
1307 ip->ip_hl = sizeof(struct ip) >> 2;
1308 ip->ip_len = tlen;
1309 ip->ip_id = 0;
1310 ip->ip_off = 0;
1311 ip->ip_sum = 0;
1312 ip->ip_p = IPPROTO_TCP;
1313 ip->ip_src = sc->sc_inc.inc_laddr;
1314 ip->ip_dst = sc->sc_inc.inc_faddr;
1315 ip->ip_ttl = sc->sc_ip_ttl;
1316 ip->ip_tos = sc->sc_ip_tos;
1317
1318 /*
1319 * See if we should do MTU discovery. Route lookups are
1320 * expensive, so we will only unset the DF bit if:
1321 *
1322 * 1) path_mtu_discovery is disabled
1323 * 2) the SCF_UNREACH flag has been set
1324 */
1325 if (path_mtu_discovery && ((sc->sc_flags & SCF_UNREACH) == 0))
1326 ip->ip_off |= IP_DF;
1327
1328 th = (struct tcphdr *)(ip + 1);
1329 }
1330 th->th_sport = sc->sc_inc.inc_lport;
1331 th->th_dport = sc->sc_inc.inc_fport;
1332
1333 th->th_seq = htonl(sc->sc_iss);
1334 th->th_ack = htonl(sc->sc_irs + 1);
1335 th->th_off = sizeof(struct tcphdr) >> 2;
1336 th->th_x2 = 0;
1337 th->th_flags = TH_SYN|TH_ACK;
1338 th->th_win = htons(sc->sc_wnd);
1339 th->th_urp = 0;
1340
1341 /* Tack on the TCP options. */
1342 if ((sc->sc_flags & SCF_NOOPT) == 0) {
1343 to.to_flags = 0;
1344
1345 to.to_mss = mssopt;
1346 to.to_flags = TOF_MSS;
1347 if (sc->sc_flags & SCF_WINSCALE) {
1348 to.to_wscale = sc->sc_requested_r_scale;
1349 to.to_flags |= TOF_SCALE;
1350 }
1351 if (sc->sc_flags & SCF_TIMESTAMP) {
1352 /* Virgin timestamp or TCP cookie enhanced one. */
1353 to.to_tsval = sc->sc_ts;
1354 to.to_tsecr = sc->sc_tsreflect;
1355 to.to_flags |= TOF_TS;
1356 }
1357 if (sc->sc_flags & SCF_SACK)
1358 to.to_flags |= TOF_SACKPERM;
1359 #ifdef TCP_SIGNATURE
1360 if (sc->sc_flags & SCF_SIGNATURE)
1361 to.to_flags |= TOF_SIGNATURE;
1362 #endif
1363 optlen = tcp_addoptions(&to, (u_char *)(th + 1));
1364
1365 /* Adjust headers by option size. */
1366 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1367 m->m_len += optlen;
1368 m->m_pkthdr.len += optlen;
1369
1370 #ifdef TCP_SIGNATURE
1371 if (sc->sc_flags & SCF_SIGNATURE)
1372 tcp_signature_compute(m, sizeof(struct ip), 0, optlen,
1373 to.to_signature, IPSEC_DIR_OUTBOUND);
1374 #endif
1375 #ifdef INET6
1376 if (sc->sc_inc.inc_isipv6)
1377 ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) + optlen);
1378 else
1379 #endif
1380 ip->ip_len += optlen;
1381 } else
1382 optlen = 0;
1383
1384 #ifdef INET6
1385 if (sc->sc_inc.inc_isipv6) {
1386 th->th_sum = 0;
1387 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen,
1388 tlen + optlen - hlen);
1389 ip6->ip6_hlim = in6_selecthlim(NULL, NULL);
1390 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
1391 } else
1392 #endif
1393 {
1394 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1395 htons(tlen + optlen - hlen + IPPROTO_TCP));
1396 m->m_pkthdr.csum_flags = CSUM_TCP;
1397 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1398 error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, NULL);
1399 }
1400 return (error);
1401 }
1402
1403 /*
1404 * The purpose of SYN cookies is to avoid keeping track of all SYN's we
1405 * receive and to be able to handle SYN floods from bogus source addresses
1406 * (where we will never receive any reply). SYN floods try to exhaust all
1407 * our memory and available slots in the SYN cache table to cause a denial
1408 * of service to legitimate users of the local host.
1409 *
1410 * The idea of SYN cookies is to encode and include all necessary information
1411 * about the connection setup state within the SYN-ACK we send back and thus
1412 * to get along without keeping any local state until the ACK to the SYN-ACK
1413 * arrives (if ever). Everything we need to know should be available from
1414 * the information we encoded in the SYN-ACK.
1415 *
1416 * More information about the theory behind SYN cookies and its first
1417 * discussion and specification can be found at:
1418 * http://cr.yp.to/syncookies.html (overview)
1419 * http://cr.yp.to/syncookies/archive (gory details)
1420 *
1421 * This implementation extends the orginal idea and first implementation
1422 * of FreeBSD by using not only the initial sequence number field to store
1423 * information but also the timestamp field if present. This way we can
1424 * keep track of the entire state we need to know to recreate the session in
1425 * its original form. Almost all TCP speakers implement RFC1323 timestamps
1426 * these days. For those that do not we still have to live with the known
1427 * shortcomings of the ISN only SYN cookies.
1428 *
1429 * Cookie layers:
1430 *
1431 * Initial sequence number we send:
1432 * 31|................................|0
1433 * DDDDDDDDDDDDDDDDDDDDDDDDDMMMRRRP
1434 * D = MD5 Digest (first dword)
1435 * M = MSS index
1436 * R = Rotation of secret
1437 * P = Odd or Even secret
1438 *
1439 * The MD5 Digest is computed with over following parameters:
1440 * a) randomly rotated secret
1441 * b) struct in_conninfo containing the remote/local ip/port (IPv4&IPv6)
1442 * c) the received initial sequence number from remote host
1443 * d) the rotation offset and odd/even bit
1444 *
1445 * Timestamp we send:
1446 * 31|................................|0
1447 * DDDDDDDDDDDDDDDDDDDDDDSSSSRRRRA5
1448 * D = MD5 Digest (third dword) (only as filler)
1449 * S = Requested send window scale
1450 * R = Requested receive window scale
1451 * A = SACK allowed
1452 * 5 = TCP-MD5 enabled (not implemented yet)
1453 * XORed with MD5 Digest (forth dword)
1454 *
1455 * The timestamp isn't cryptographically secure and doesn't need to be.
1456 * The double use of the MD5 digest dwords ties it to a specific remote/
1457 * local host/port, remote initial sequence number and our local time
1458 * limited secret. A received timestamp is reverted (XORed) and then
1459 * the contained MD5 dword is compared to the computed one to ensure the
1460 * timestamp belongs to the SYN-ACK we sent. The other parameters may
1461 * have been tampered with but this isn't different from supplying bogus
1462 * values in the SYN in the first place.
1463 *
1464 * Some problems with SYN cookies remain however:
1465 * Consider the problem of a recreated (and retransmitted) cookie. If the
1466 * original SYN was accepted, the connection is established. The second
1467 * SYN is inflight, and if it arrives with an ISN that falls within the
1468 * receive window, the connection is killed.
1469 *
1470 * Notes:
1471 * A heuristic to determine when to accept syn cookies is not necessary.
1472 * An ACK flood would cause the syncookie verification to be attempted,
1473 * but a SYN flood causes syncookies to be generated. Both are of equal
1474 * cost, so there's no point in trying to optimize the ACK flood case.
1475 * Also, if you don't process certain ACKs for some reason, then all someone
1476 * would have to do is launch a SYN and ACK flood at the same time, which
1477 * would stop cookie verification and defeat the entire purpose of syncookies.
1478 */
1479 static int tcp_sc_msstab[] = { 0, 256, 468, 536, 996, 1452, 1460, 8960 };
1480
1481 static void
1482 syncookie_generate(struct syncache_head *sch, struct syncache *sc,
1483 u_int32_t *flowlabel)
1484 {
1485 MD5_CTX ctx;
1486 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)];
1487 u_int32_t data;
1488 u_int32_t *secbits;
1489 u_int off, pmss, mss;
1490 int i;
1491
1492 SCH_LOCK_ASSERT(sch);
1493
1494 /* Which of the two secrets to use. */
1495 secbits = sch->sch_oddeven ?
1496 sch->sch_secbits_odd : sch->sch_secbits_even;
1497
1498 /* Reseed secret if too old. */
1499 if (sch->sch_reseed < time_uptime) {
1500 sch->sch_oddeven = sch->sch_oddeven ? 0 : 1; /* toggle */
1501 secbits = sch->sch_oddeven ?
1502 sch->sch_secbits_odd : sch->sch_secbits_even;
1503 for (i = 0; i < SYNCOOKIE_SECRET_SIZE; i++)
1504 secbits[i] = arc4random();
1505 sch->sch_reseed = time_uptime + SYNCOOKIE_LIFETIME;
1506 }
1507
1508 /* Secret rotation offset. */
1509 off = sc->sc_iss & 0x7; /* iss was randomized before */
1510
1511 /* Maximum segment size calculation. */
1512 pmss = max( min(sc->sc_peer_mss, tcp_mssopt(&sc->sc_inc)), tcp_minmss);
1513 for (mss = sizeof(tcp_sc_msstab) / sizeof(int) - 1; mss > 0; mss--)
1514 if (tcp_sc_msstab[mss] <= pmss)
1515 break;
1516
1517 /* Fold parameters and MD5 digest into the ISN we will send. */
1518 data = sch->sch_oddeven;/* odd or even secret, 1 bit */
1519 data |= off << 1; /* secret offset, derived from iss, 3 bits */
1520 data |= mss << 4; /* mss, 3 bits */
1521
1522 MD5Init(&ctx);
1523 MD5Update(&ctx, ((u_int8_t *)secbits) + off,
1524 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off);
1525 MD5Update(&ctx, secbits, off);
1526 MD5Update(&ctx, &sc->sc_inc, sizeof(sc->sc_inc));
1527 MD5Update(&ctx, &sc->sc_irs, sizeof(sc->sc_irs));
1528 MD5Update(&ctx, &data, sizeof(data));
1529 MD5Final((u_int8_t *)&md5_buffer, &ctx);
1530
1531 data |= (md5_buffer[0] << 7);
1532 sc->sc_iss = data;
1533
1534 #ifdef INET6
1535 *flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
1536 #endif
1537
1538 /* Additional parameters are stored in the timestamp if present. */
1539 if (sc->sc_flags & SCF_TIMESTAMP) {
1540 data = ((sc->sc_flags & SCF_SIGNATURE) ? 1 : 0); /* TCP-MD5, 1 bit */
1541 data |= ((sc->sc_flags & SCF_SACK) ? 1 : 0) << 1; /* SACK, 1 bit */
1542 data |= sc->sc_requested_s_scale << 2; /* SWIN scale, 4 bits */
1543 data |= sc->sc_requested_r_scale << 6; /* RWIN scale, 4 bits */
1544 data |= md5_buffer[2] << 10; /* more digest bits */
1545 data ^= md5_buffer[3];
1546 sc->sc_ts = data;
1547 sc->sc_tsoff = data - ticks; /* after XOR */
1548 }
1549
1550 tcpstat.tcps_sc_sendcookie++;
1551 return;
1552 }
1553
1554 static struct syncache *
1555 syncookie_lookup(struct in_conninfo *inc, struct syncache_head *sch,
1556 struct syncache *sc, struct tcpopt *to, struct tcphdr *th,
1557 struct socket *so)
1558 {
1559 MD5_CTX ctx;
1560 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)];
1561 u_int32_t data = 0;
1562 u_int32_t *secbits;
1563 tcp_seq ack, seq;
1564 int off, mss, wnd, flags;
1565
1566 SCH_LOCK_ASSERT(sch);
1567
1568 /*
1569 * Pull information out of SYN-ACK/ACK and
1570 * revert sequence number advances.
1571 */
1572 ack = th->th_ack - 1;
1573 seq = th->th_seq - 1;
1574 off = (ack >> 1) & 0x7;
1575 mss = (ack >> 4) & 0x7;
1576 flags = ack & 0x7f;
1577
1578 /* Which of the two secrets to use. */
1579 secbits = (flags & 0x1) ? sch->sch_secbits_odd : sch->sch_secbits_even;
1580
1581 /*
1582 * The secret wasn't updated for the lifetime of a syncookie,
1583 * so this SYN-ACK/ACK is either too old (replay) or totally bogus.
1584 */
1585 if (sch->sch_reseed < time_uptime) {
1586 return (NULL);
1587 }
1588
1589 /* Recompute the digest so we can compare it. */
1590 MD5Init(&ctx);
1591 MD5Update(&ctx, ((u_int8_t *)secbits) + off,
1592 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off);
1593 MD5Update(&ctx, secbits, off);
1594 MD5Update(&ctx, inc, sizeof(*inc));
1595 MD5Update(&ctx, &seq, sizeof(seq));
1596 MD5Update(&ctx, &flags, sizeof(flags));
1597 MD5Final((u_int8_t *)&md5_buffer, &ctx);
1598
1599 /* Does the digest part of or ACK'ed ISS match? */
1600 if ((ack & (~0x7f)) != (md5_buffer[0] << 7))
1601 return (NULL);
1602
1603 /* Does the digest part of our reflected timestamp match? */
1604 if (to->to_flags & TOF_TS) {
1605 data = md5_buffer[3] ^ to->to_tsecr;
1606 if ((data & (~0x3ff)) != (md5_buffer[2] << 10))
1607 return (NULL);
1608 }
1609
1610 /* Fill in the syncache values. */
1611 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo));
1612 sc->sc_ipopts = NULL;
1613
1614 sc->sc_irs = seq;
1615 sc->sc_iss = ack;
1616
1617 #ifdef INET6
1618 if (inc->inc_isipv6) {
1619 if (sotoinpcb(so)->in6p_flags & IN6P_AUTOFLOWLABEL)
1620 sc->sc_flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
1621 } else
1622 #endif
1623 {
1624 sc->sc_ip_ttl = sotoinpcb(so)->inp_ip_ttl;
1625 sc->sc_ip_tos = sotoinpcb(so)->inp_ip_tos;
1626 }
1627
1628 /* Additional parameters that were encoded in the timestamp. */
1629 if (data) {
1630 sc->sc_flags |= SCF_TIMESTAMP;
1631 sc->sc_tsreflect = to->to_tsval;
1632 sc->sc_ts = to->to_tsecr;
1633 sc->sc_tsoff = to->to_tsecr - ticks;
1634 sc->sc_flags |= (data & 0x1) ? SCF_SIGNATURE : 0;
1635 sc->sc_flags |= ((data >> 1) & 0x1) ? SCF_SACK : 0;
1636 sc->sc_requested_s_scale = min((data >> 2) & 0xf,
1637 TCP_MAX_WINSHIFT);
1638 sc->sc_requested_r_scale = min((data >> 6) & 0xf,
1639 TCP_MAX_WINSHIFT);
1640 if (sc->sc_requested_s_scale || sc->sc_requested_r_scale)
1641 sc->sc_flags |= SCF_WINSCALE;
1642 } else
1643 sc->sc_flags |= SCF_NOOPT;
1644
1645 wnd = sbspace(&so->so_rcv);
1646 wnd = imax(wnd, 0);
1647 wnd = imin(wnd, TCP_MAXWIN);
1648 sc->sc_wnd = wnd;
1649
1650 sc->sc_rxmits = 0;
1651 sc->sc_peer_mss = tcp_sc_msstab[mss];
1652
1653 tcpstat.tcps_sc_recvcookie++;
1654 return (sc);
1655 }
1656
1657 /*
1658 * Returns the current number of syncache entries. This number
1659 * will probably change before you get around to calling
1660 * syncache_pcblist.
1661 */
1662
1663 int
1664 syncache_pcbcount(void)
1665 {
1666 struct syncache_head *sch;
1667 int count, i;
1668
1669 for (count = 0, i = 0; i < tcp_syncache.hashsize; i++) {
1670 /* No need to lock for a read. */
1671 sch = &tcp_syncache.hashbase[i];
1672 count += sch->sch_length;
1673 }
1674 return count;
1675 }
1676
1677 /*
1678 * Exports the syncache entries to userland so that netstat can display
1679 * them alongside the other sockets. This function is intended to be
1680 * called only from tcp_pcblist.
1681 *
1682 * Due to concurrency on an active system, the number of pcbs exported
1683 * may have no relation to max_pcbs. max_pcbs merely indicates the
1684 * amount of space the caller allocated for this function to use.
1685 */
1686 int
1687 syncache_pcblist(struct sysctl_req *req, int max_pcbs, int *pcbs_exported)
1688 {
1689 struct xtcpcb xt;
1690 struct syncache *sc;
1691 struct syncache_head *sch;
1692 int count, error, i;
1693
1694 for (count = 0, error = 0, i = 0; i < tcp_syncache.hashsize; i++) {
1695 sch = &tcp_syncache.hashbase[i];
1696 SCH_LOCK(sch);
1697 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
1698 if (count >= max_pcbs) {
1699 SCH_UNLOCK(sch);
1700 goto exit;
1701 }
1702 bzero(&xt, sizeof(xt));
1703 xt.xt_len = sizeof(xt);
1704 if (sc->sc_inc.inc_isipv6)
1705 xt.xt_inp.inp_vflag = INP_IPV6;
1706 else
1707 xt.xt_inp.inp_vflag = INP_IPV4;
1708 bcopy(&sc->sc_inc, &xt.xt_inp.inp_inc, sizeof (struct in_conninfo));
1709 xt.xt_tp.t_inpcb = &xt.xt_inp;
1710 xt.xt_tp.t_state = TCPS_SYN_RECEIVED;
1711 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1712 xt.xt_socket.xso_len = sizeof (struct xsocket);
1713 xt.xt_socket.so_type = SOCK_STREAM;
1714 xt.xt_socket.so_state = SS_ISCONNECTING;
1715 error = SYSCTL_OUT(req, &xt, sizeof xt);
1716 if (error) {
1717 SCH_UNLOCK(sch);
1718 goto exit;
1719 }
1720 count++;
1721 }
1722 SCH_UNLOCK(sch);
1723 }
1724 exit:
1725 *pcbs_exported = count;
1726 return error;
1727 }
1728
Cache object: 20c9742372b77635f8adb2589441fe99
|