2 * Copyright (c) 2003, 2004 Jeffrey M. Hsu. All rights reserved.
3 * Copyright (c) 2003, 2004 The DragonFly Project. All rights reserved.
5 * This code is derived from software contributed to The DragonFly Project
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. Neither the name of The DragonFly Project nor the names of its
17 * contributors may be used to endorse or promote products derived
18 * from this software without specific, prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * All advertising materials mentioning features or use of this software
36 * must display the following acknowledgement:
37 * This product includes software developed by Jeffrey M. Hsu.
39 * Copyright (c) 2001 Networks Associates Technologies, Inc.
40 * All rights reserved.
42 * This software was developed for the FreeBSD Project by Jonathan Lemon
43 * and NAI Labs, the Security Research Division of Network Associates, Inc.
44 * under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the
45 * DARPA CHATS research program.
47 * Redistribution and use in source and binary forms, with or without
48 * modification, are permitted provided that the following conditions
50 * 1. Redistributions of source code must retain the above copyright
51 * notice, this list of conditions and the following disclaimer.
52 * 2. Redistributions in binary form must reproduce the above copyright
53 * notice, this list of conditions and the following disclaimer in the
54 * documentation and/or other materials provided with the distribution.
55 * 3. The name of the author may not be used to endorse or promote
56 * products derived from this software without specific prior written
59 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
60 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
61 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
62 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
63 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
64 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
65 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
66 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
67 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
68 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
71 * $FreeBSD: src/sys/netinet/tcp_syncache.c,v 1.5.2.14 2003/02/24 04:02:27 silby Exp $
72 * $DragonFly: src/sys/netinet/tcp_syncache.c,v 1.29 2007/04/22 01:13:14 dillon Exp $
75 #include "opt_inet6.h"
76 #include "opt_ipsec.h"
78 #include <sys/param.h>
79 #include <sys/systm.h>
80 #include <sys/kernel.h>
81 #include <sys/sysctl.h>
82 #include <sys/malloc.h>
85 #include <sys/proc.h> /* for proc0 declaration */
86 #include <sys/random.h>
87 #include <sys/socket.h>
88 #include <sys/socketvar.h>
89 #include <sys/in_cksum.h>
91 #include <sys/msgport2.h>
94 #include <net/route.h>
96 #include <netinet/in.h>
97 #include <netinet/in_systm.h>
98 #include <netinet/ip.h>
99 #include <netinet/in_var.h>
100 #include <netinet/in_pcb.h>
101 #include <netinet/ip_var.h>
102 #include <netinet/ip6.h>
104 #include <netinet/icmp6.h>
105 #include <netinet6/nd6.h>
107 #include <netinet6/ip6_var.h>
108 #include <netinet6/in6_pcb.h>
109 #include <netinet/tcp.h>
110 #include <netinet/tcp_fsm.h>
111 #include <netinet/tcp_seq.h>
112 #include <netinet/tcp_timer.h>
113 #include <netinet/tcp_var.h>
114 #include <netinet6/tcp6_var.h>
117 #include <netinet6/ipsec.h>
119 #include <netinet6/ipsec6.h>
121 #include <netproto/key/key.h>
125 #include <netproto/ipsec/ipsec.h>
127 #include <netproto/ipsec/ipsec6.h>
129 #include <netproto/ipsec/key.h>
131 #endif /*FAST_IPSEC*/
133 #include <vm/vm_zone.h>
135 static int tcp_syncookies
= 1;
136 SYSCTL_INT(_net_inet_tcp
, OID_AUTO
, syncookies
, CTLFLAG_RW
,
138 "Use TCP SYN cookies if the syncache overflows");
140 static void syncache_drop(struct syncache
*, struct syncache_head
*);
141 static void syncache_free(struct syncache
*);
142 static void syncache_insert(struct syncache
*, struct syncache_head
*);
143 struct syncache
*syncache_lookup(struct in_conninfo
*, struct syncache_head
**);
144 static int syncache_respond(struct syncache
*, struct mbuf
*);
145 static struct socket
*syncache_socket(struct syncache
*, struct socket
*);
146 static void syncache_timer(void *);
147 static u_int32_t
syncookie_generate(struct syncache
*);
148 static struct syncache
*syncookie_lookup(struct in_conninfo
*,
149 struct tcphdr
*, struct socket
*);
152 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies.
153 * 3 retransmits corresponds to a timeout of (1 + 2 + 4 + 8 == 15) seconds,
154 * the odds are that the user has given up attempting to connect by then.
156 #define SYNCACHE_MAXREXMTS 3
158 /* Arbitrary values */
159 #define TCP_SYNCACHE_HASHSIZE 512
160 #define TCP_SYNCACHE_BUCKETLIMIT 30
162 struct netmsg_sc_timer
{
163 struct lwkt_msg nm_lmsg
;
164 struct msgrec
*nm_mrec
; /* back pointer to containing msgrec */
168 struct netmsg_sc_timer msg
;
169 lwkt_port_t port
; /* constant after init */
170 int slot
; /* constant after init */
173 static int syncache_timer_handler(lwkt_msg_t
);
175 struct tcp_syncache
{
176 struct vm_zone
*zone
;
184 static struct tcp_syncache tcp_syncache
;
186 struct tcp_syncache_percpu
{
187 struct syncache_head
*hashbase
;
189 TAILQ_HEAD(, syncache
) timerq
[SYNCACHE_MAXREXMTS
+ 1];
190 struct callout tt_timerq
[SYNCACHE_MAXREXMTS
+ 1];
191 struct msgrec mrec
[SYNCACHE_MAXREXMTS
+ 1];
193 static struct tcp_syncache_percpu tcp_syncache_percpu
[MAXCPU
];
195 static struct lwkt_port syncache_null_rport
;
197 SYSCTL_NODE(_net_inet_tcp
, OID_AUTO
, syncache
, CTLFLAG_RW
, 0, "TCP SYN cache");
199 SYSCTL_INT(_net_inet_tcp_syncache
, OID_AUTO
, bucketlimit
, CTLFLAG_RD
,
200 &tcp_syncache
.bucket_limit
, 0, "Per-bucket hash limit for syncache");
202 SYSCTL_INT(_net_inet_tcp_syncache
, OID_AUTO
, cachelimit
, CTLFLAG_RD
,
203 &tcp_syncache
.cache_limit
, 0, "Overall entry limit for syncache");
207 SYSCTL_INT(_net_inet_tcp_syncache
, OID_AUTO
, count
, CTLFLAG_RD
,
208 &tcp_syncache
.cache_count
, 0, "Current number of entries in syncache");
211 SYSCTL_INT(_net_inet_tcp_syncache
, OID_AUTO
, hashsize
, CTLFLAG_RD
,
212 &tcp_syncache
.hashsize
, 0, "Size of TCP syncache hashtable");
214 SYSCTL_INT(_net_inet_tcp_syncache
, OID_AUTO
, rexmtlimit
, CTLFLAG_RW
,
215 &tcp_syncache
.rexmt_limit
, 0, "Limit on SYN/ACK retransmissions");
217 static MALLOC_DEFINE(M_SYNCACHE
, "syncache", "TCP syncache");
219 #define SYNCACHE_HASH(inc, mask) \
220 ((tcp_syncache.hash_secret ^ \
221 (inc)->inc_faddr.s_addr ^ \
222 ((inc)->inc_faddr.s_addr >> 16) ^ \
223 (inc)->inc_fport ^ (inc)->inc_lport) & mask)
225 #define SYNCACHE_HASH6(inc, mask) \
226 ((tcp_syncache.hash_secret ^ \
227 (inc)->inc6_faddr.s6_addr32[0] ^ \
228 (inc)->inc6_faddr.s6_addr32[3] ^ \
229 (inc)->inc_fport ^ (inc)->inc_lport) & mask)
231 #define ENDPTS_EQ(a, b) ( \
232 (a)->ie_fport == (b)->ie_fport && \
233 (a)->ie_lport == (b)->ie_lport && \
234 (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr && \
235 (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr \
238 #define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0)
241 syncache_timeout(struct tcp_syncache_percpu
*syncache_percpu
,
242 struct syncache
*sc
, int slot
)
244 sc
->sc_rxtslot
= slot
;
245 sc
->sc_rxttime
= ticks
+ TCPTV_RTOBASE
* tcp_backoff
[slot
];
246 TAILQ_INSERT_TAIL(&syncache_percpu
->timerq
[slot
], sc
, sc_timerq
);
247 if (!callout_active(&syncache_percpu
->tt_timerq
[slot
])) {
248 callout_reset(&syncache_percpu
->tt_timerq
[slot
],
249 TCPTV_RTOBASE
* tcp_backoff
[slot
],
251 &syncache_percpu
->mrec
[slot
]);
256 syncache_free(struct syncache
*sc
)
260 const boolean_t isipv6
= sc
->sc_inc
.inc_isipv6
;
262 const boolean_t isipv6
= FALSE
;
266 m_free(sc
->sc_ipopts
);
268 rt
= isipv6
? sc
->sc_route6
.ro_rt
: sc
->sc_route
.ro_rt
;
271 * If this is the only reference to a protocol-cloned
272 * route, remove it immediately.
274 if ((rt
->rt_flags
& RTF_WASCLONED
) && rt
->rt_refcnt
== 1)
275 rtrequest(RTM_DELETE
, rt_key(rt
), rt
->rt_gateway
,
276 rt_mask(rt
), rt
->rt_flags
, NULL
);
280 zfree(tcp_syncache
.zone
, sc
);
288 tcp_syncache
.hashsize
= TCP_SYNCACHE_HASHSIZE
;
289 tcp_syncache
.bucket_limit
= TCP_SYNCACHE_BUCKETLIMIT
;
290 tcp_syncache
.cache_limit
=
291 tcp_syncache
.hashsize
* tcp_syncache
.bucket_limit
;
292 tcp_syncache
.rexmt_limit
= SYNCACHE_MAXREXMTS
;
293 tcp_syncache
.hash_secret
= karc4random();
295 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize",
296 &tcp_syncache
.hashsize
);
297 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit",
298 &tcp_syncache
.cache_limit
);
299 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit",
300 &tcp_syncache
.bucket_limit
);
301 if (!powerof2(tcp_syncache
.hashsize
)) {
302 kprintf("WARNING: syncache hash size is not a power of 2.\n");
303 tcp_syncache
.hashsize
= 512; /* safe default */
305 tcp_syncache
.hashmask
= tcp_syncache
.hashsize
- 1;
307 lwkt_initport_null_rport(&syncache_null_rport
, NULL
);
309 for (cpu
= 0; cpu
< ncpus2
; cpu
++) {
310 struct tcp_syncache_percpu
*syncache_percpu
;
312 syncache_percpu
= &tcp_syncache_percpu
[cpu
];
313 /* Allocate the hash table. */
314 MALLOC(syncache_percpu
->hashbase
, struct syncache_head
*,
315 tcp_syncache
.hashsize
* sizeof(struct syncache_head
),
316 M_SYNCACHE
, M_WAITOK
);
318 /* Initialize the hash buckets. */
319 for (i
= 0; i
< tcp_syncache
.hashsize
; i
++) {
320 struct syncache_head
*bucket
;
322 bucket
= &syncache_percpu
->hashbase
[i
];
323 TAILQ_INIT(&bucket
->sch_bucket
);
324 bucket
->sch_length
= 0;
327 for (i
= 0; i
<= SYNCACHE_MAXREXMTS
; i
++) {
328 /* Initialize the timer queues. */
329 TAILQ_INIT(&syncache_percpu
->timerq
[i
]);
330 callout_init(&syncache_percpu
->tt_timerq
[i
]);
332 syncache_percpu
->mrec
[i
].slot
= i
;
333 syncache_percpu
->mrec
[i
].port
= tcp_cport(cpu
);
334 syncache_percpu
->mrec
[i
].msg
.nm_mrec
=
335 &syncache_percpu
->mrec
[i
];
336 lwkt_initmsg(&syncache_percpu
->mrec
[i
].msg
.nm_lmsg
,
337 &syncache_null_rport
, 0,
338 lwkt_cmd_func(syncache_timer_handler
),
344 * Allocate the syncache entries. Allow the zone to allocate one
345 * more entry than cache limit, so a new entry can bump out an
348 tcp_syncache
.zone
= zinit("syncache", sizeof(struct syncache
),
349 tcp_syncache
.cache_limit
* ncpus2
, ZONE_INTERRUPT
, 0);
350 tcp_syncache
.cache_limit
-= 1;
354 syncache_insert(struct syncache
*sc
, struct syncache_head
*sch
)
356 struct tcp_syncache_percpu
*syncache_percpu
;
357 struct syncache
*sc2
;
360 syncache_percpu
= &tcp_syncache_percpu
[mycpu
->gd_cpuid
];
363 * Make sure that we don't overflow the per-bucket
364 * limit or the total cache size limit.
366 if (sch
->sch_length
>= tcp_syncache
.bucket_limit
) {
368 * The bucket is full, toss the oldest element.
370 sc2
= TAILQ_FIRST(&sch
->sch_bucket
);
371 sc2
->sc_tp
->ts_recent
= ticks
;
372 syncache_drop(sc2
, sch
);
373 tcpstat
.tcps_sc_bucketoverflow
++;
374 } else if (syncache_percpu
->cache_count
>= tcp_syncache
.cache_limit
) {
376 * The cache is full. Toss the oldest entry in the
377 * entire cache. This is the front entry in the
378 * first non-empty timer queue with the largest
381 for (i
= SYNCACHE_MAXREXMTS
; i
>= 0; i
--) {
382 sc2
= TAILQ_FIRST(&syncache_percpu
->timerq
[i
]);
386 sc2
->sc_tp
->ts_recent
= ticks
;
387 syncache_drop(sc2
, NULL
);
388 tcpstat
.tcps_sc_cacheoverflow
++;
391 /* Initialize the entry's timer. */
392 syncache_timeout(syncache_percpu
, sc
, 0);
394 /* Put it into the bucket. */
395 TAILQ_INSERT_TAIL(&sch
->sch_bucket
, sc
, sc_hash
);
397 syncache_percpu
->cache_count
++;
398 tcpstat
.tcps_sc_added
++;
402 syncache_drop(struct syncache
*sc
, struct syncache_head
*sch
)
404 struct tcp_syncache_percpu
*syncache_percpu
;
406 const boolean_t isipv6
= sc
->sc_inc
.inc_isipv6
;
408 const boolean_t isipv6
= FALSE
;
411 syncache_percpu
= &tcp_syncache_percpu
[mycpu
->gd_cpuid
];
415 sch
= &syncache_percpu
->hashbase
[
416 SYNCACHE_HASH6(&sc
->sc_inc
, tcp_syncache
.hashmask
)];
418 sch
= &syncache_percpu
->hashbase
[
419 SYNCACHE_HASH(&sc
->sc_inc
, tcp_syncache
.hashmask
)];
423 TAILQ_REMOVE(&sch
->sch_bucket
, sc
, sc_hash
);
425 syncache_percpu
->cache_count
--;
428 * Remove the entry from the syncache timer/timeout queue. Note
429 * that we do not try to stop any running timer since we do not know
430 * whether the timer's message is in-transit or not. Since timeouts
431 * are fairly long, taking an unneeded callout does not detrimentally
432 * effect performance.
434 TAILQ_REMOVE(&syncache_percpu
->timerq
[sc
->sc_rxtslot
], sc
, sc_timerq
);
440 * Place a timeout message on the TCP thread's message queue.
441 * This routine runs in soft interrupt context.
443 * An invariant is for this routine to be called, the callout must
444 * have been active. Note that the callout is not deactivated until
445 * after the message has been processed in syncache_timer_handler() below.
448 syncache_timer(void *p
)
450 struct netmsg_sc_timer
*msg
= p
;
452 lwkt_sendmsg(msg
->nm_mrec
->port
, &msg
->nm_lmsg
);
456 * Service a timer message queued by timer expiration.
457 * This routine runs in the TCP protocol thread.
459 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
460 * If we have retransmitted an entry the maximum number of times, expire it.
462 * When we finish processing timed-out entries, we restart the timer if there
463 * are any entries still on the queue and deactivate it otherwise. Only after
464 * a timer has been deactivated here can it be restarted by syncache_timeout().
467 syncache_timer_handler(lwkt_msg_t msg
)
469 struct tcp_syncache_percpu
*syncache_percpu
;
470 struct syncache
*sc
, *nsc
;
474 slot
= ((struct netmsg_sc_timer
*)msg
)->nm_mrec
->slot
;
475 syncache_percpu
= &tcp_syncache_percpu
[mycpu
->gd_cpuid
];
477 nsc
= TAILQ_FIRST(&syncache_percpu
->timerq
[slot
]);
478 while (nsc
!= NULL
) {
479 if (ticks
< nsc
->sc_rxttime
)
480 break; /* finished because timerq sorted by time */
482 inp
= sc
->sc_tp
->t_inpcb
;
483 if (slot
== SYNCACHE_MAXREXMTS
||
484 slot
>= tcp_syncache
.rexmt_limit
||
485 inp
->inp_gencnt
!= sc
->sc_inp_gencnt
) {
486 nsc
= TAILQ_NEXT(sc
, sc_timerq
);
487 syncache_drop(sc
, NULL
);
488 tcpstat
.tcps_sc_stale
++;
492 * syncache_respond() may call back into the syncache to
493 * to modify another entry, so do not obtain the next
494 * entry on the timer chain until it has completed.
496 syncache_respond(sc
, NULL
);
497 nsc
= TAILQ_NEXT(sc
, sc_timerq
);
498 tcpstat
.tcps_sc_retransmitted
++;
499 TAILQ_REMOVE(&syncache_percpu
->timerq
[slot
], sc
, sc_timerq
);
500 syncache_timeout(syncache_percpu
, sc
, slot
+ 1);
503 callout_reset(&syncache_percpu
->tt_timerq
[slot
],
504 nsc
->sc_rxttime
- ticks
, syncache_timer
,
505 &syncache_percpu
->mrec
[slot
]);
507 callout_deactivate(&syncache_percpu
->tt_timerq
[slot
]);
509 lwkt_replymsg(msg
, 0);
514 * Find an entry in the syncache.
517 syncache_lookup(struct in_conninfo
*inc
, struct syncache_head
**schp
)
519 struct tcp_syncache_percpu
*syncache_percpu
;
521 struct syncache_head
*sch
;
523 syncache_percpu
= &tcp_syncache_percpu
[mycpu
->gd_cpuid
];
525 if (inc
->inc_isipv6
) {
526 sch
= &syncache_percpu
->hashbase
[
527 SYNCACHE_HASH6(inc
, tcp_syncache
.hashmask
)];
529 TAILQ_FOREACH(sc
, &sch
->sch_bucket
, sc_hash
)
530 if (ENDPTS6_EQ(&inc
->inc_ie
, &sc
->sc_inc
.inc_ie
))
535 sch
= &syncache_percpu
->hashbase
[
536 SYNCACHE_HASH(inc
, tcp_syncache
.hashmask
)];
538 TAILQ_FOREACH(sc
, &sch
->sch_bucket
, sc_hash
) {
540 if (sc
->sc_inc
.inc_isipv6
)
543 if (ENDPTS_EQ(&inc
->inc_ie
, &sc
->sc_inc
.inc_ie
))
551 * This function is called when we get a RST for a
552 * non-existent connection, so that we can see if the
553 * connection is in the syn cache. If it is, zap it.
556 syncache_chkrst(struct in_conninfo
*inc
, struct tcphdr
*th
)
559 struct syncache_head
*sch
;
561 sc
= syncache_lookup(inc
, &sch
);
565 * If the RST bit is set, check the sequence number to see
566 * if this is a valid reset segment.
568 * In all states except SYN-SENT, all reset (RST) segments
569 * are validated by checking their SEQ-fields. A reset is
570 * valid if its sequence number is in the window.
572 * The sequence number in the reset segment is normally an
573 * echo of our outgoing acknowlegement numbers, but some hosts
574 * send a reset with the sequence number at the rightmost edge
575 * of our receive window, and we have to handle this case.
577 if (SEQ_GEQ(th
->th_seq
, sc
->sc_irs
) &&
578 SEQ_LEQ(th
->th_seq
, sc
->sc_irs
+ sc
->sc_wnd
)) {
579 syncache_drop(sc
, sch
);
580 tcpstat
.tcps_sc_reset
++;
585 syncache_badack(struct in_conninfo
*inc
)
588 struct syncache_head
*sch
;
590 sc
= syncache_lookup(inc
, &sch
);
592 syncache_drop(sc
, sch
);
593 tcpstat
.tcps_sc_badack
++;
598 syncache_unreach(struct in_conninfo
*inc
, struct tcphdr
*th
)
601 struct syncache_head
*sch
;
603 /* we are called at splnet() here */
604 sc
= syncache_lookup(inc
, &sch
);
608 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */
609 if (ntohl(th
->th_seq
) != sc
->sc_iss
)
613 * If we've rertransmitted 3 times and this is our second error,
614 * we remove the entry. Otherwise, we allow it to continue on.
615 * This prevents us from incorrectly nuking an entry during a
616 * spurious network outage.
620 if ((sc
->sc_flags
& SCF_UNREACH
) == 0 || sc
->sc_rxtslot
< 3) {
621 sc
->sc_flags
|= SCF_UNREACH
;
624 syncache_drop(sc
, sch
);
625 tcpstat
.tcps_sc_unreach
++;
629 * Build a new TCP socket structure from a syncache entry.
631 static struct socket
*
632 syncache_socket(struct syncache
*sc
, struct socket
*lso
)
634 struct inpcb
*inp
= NULL
, *linp
;
638 const boolean_t isipv6
= sc
->sc_inc
.inc_isipv6
;
640 const boolean_t isipv6
= FALSE
;
644 * Ok, create the full blown connection, and set things up
645 * as they would have been set up if we had created the
646 * connection when the SYN arrived. If we can't create
647 * the connection, abort it.
649 so
= sonewconn(lso
, SS_ISCONNECTED
);
652 * Drop the connection; we will send a RST if the peer
653 * retransmits the ACK,
655 tcpstat
.tcps_listendrop
++;
662 * Insert new socket into hash list.
664 inp
->inp_inc
.inc_isipv6
= sc
->sc_inc
.inc_isipv6
;
666 inp
->in6p_laddr
= sc
->sc_inc
.inc6_laddr
;
669 inp
->inp_vflag
&= ~INP_IPV6
;
670 inp
->inp_vflag
|= INP_IPV4
;
672 inp
->inp_laddr
= sc
->sc_inc
.inc_laddr
;
674 inp
->inp_lport
= sc
->sc_inc
.inc_lport
;
675 if (in_pcbinsporthash(inp
) != 0) {
677 * Undo the assignments above if we failed to
678 * put the PCB on the hash lists.
681 inp
->in6p_laddr
= kin6addr_any
;
683 inp
->inp_laddr
.s_addr
= INADDR_ANY
;
689 /* copy old policy into new socket's */
690 if (ipsec_copy_policy(linp
->inp_sp
, inp
->inp_sp
))
691 kprintf("syncache_expand: could not copy policy\n");
694 struct in6_addr laddr6
;
695 struct sockaddr_in6 sin6
;
697 * Inherit socket options from the listening socket.
698 * Note that in6p_inputopts are not (and should not be)
699 * copied, since it stores previously received options and is
700 * used to detect if each new option is different than the
701 * previous one and hence should be passed to a user.
702 * If we copied in6p_inputopts, a user would not be able to
703 * receive options just after calling the accept system call.
705 inp
->inp_flags
|= linp
->inp_flags
& INP_CONTROLOPTS
;
706 if (linp
->in6p_outputopts
)
707 inp
->in6p_outputopts
=
708 ip6_copypktopts(linp
->in6p_outputopts
, M_INTWAIT
);
709 inp
->in6p_route
= sc
->sc_route6
;
710 sc
->sc_route6
.ro_rt
= NULL
;
712 sin6
.sin6_family
= AF_INET6
;
713 sin6
.sin6_len
= sizeof sin6
;
714 sin6
.sin6_addr
= sc
->sc_inc
.inc6_faddr
;
715 sin6
.sin6_port
= sc
->sc_inc
.inc_fport
;
716 sin6
.sin6_flowinfo
= sin6
.sin6_scope_id
= 0;
717 laddr6
= inp
->in6p_laddr
;
718 if (IN6_IS_ADDR_UNSPECIFIED(&inp
->in6p_laddr
))
719 inp
->in6p_laddr
= sc
->sc_inc
.inc6_laddr
;
720 if (in6_pcbconnect(inp
, (struct sockaddr
*)&sin6
, &thread0
)) {
721 inp
->in6p_laddr
= laddr6
;
725 struct in_addr laddr
;
726 struct sockaddr_in sin
;
728 inp
->inp_options
= ip_srcroute();
729 if (inp
->inp_options
== NULL
) {
730 inp
->inp_options
= sc
->sc_ipopts
;
731 sc
->sc_ipopts
= NULL
;
733 inp
->inp_route
= sc
->sc_route
;
734 sc
->sc_route
.ro_rt
= NULL
;
736 sin
.sin_family
= AF_INET
;
737 sin
.sin_len
= sizeof sin
;
738 sin
.sin_addr
= sc
->sc_inc
.inc_faddr
;
739 sin
.sin_port
= sc
->sc_inc
.inc_fport
;
740 bzero(sin
.sin_zero
, sizeof sin
.sin_zero
);
741 laddr
= inp
->inp_laddr
;
742 if (inp
->inp_laddr
.s_addr
== INADDR_ANY
)
743 inp
->inp_laddr
= sc
->sc_inc
.inc_laddr
;
744 if (in_pcbconnect(inp
, (struct sockaddr
*)&sin
, &thread0
)) {
745 inp
->inp_laddr
= laddr
;
751 tp
->t_state
= TCPS_SYN_RECEIVED
;
752 tp
->iss
= sc
->sc_iss
;
753 tp
->irs
= sc
->sc_irs
;
756 tp
->snd_wl1
= sc
->sc_irs
;
757 tp
->rcv_up
= sc
->sc_irs
+ 1;
758 tp
->rcv_wnd
= sc
->sc_wnd
;
759 tp
->rcv_adv
+= tp
->rcv_wnd
;
761 tp
->t_flags
= sototcpcb(lso
)->t_flags
& (TF_NOPUSH
| TF_NODELAY
);
762 if (sc
->sc_flags
& SCF_NOOPT
)
763 tp
->t_flags
|= TF_NOOPT
;
764 if (sc
->sc_flags
& SCF_WINSCALE
) {
765 tp
->t_flags
|= TF_REQ_SCALE
| TF_RCVD_SCALE
;
766 tp
->requested_s_scale
= sc
->sc_requested_s_scale
;
767 tp
->request_r_scale
= sc
->sc_request_r_scale
;
769 if (sc
->sc_flags
& SCF_TIMESTAMP
) {
770 tp
->t_flags
|= TF_REQ_TSTMP
| TF_RCVD_TSTMP
;
771 tp
->ts_recent
= sc
->sc_tsrecent
;
772 tp
->ts_recent_age
= ticks
;
774 if (sc
->sc_flags
& SCF_CC
) {
776 * Initialization of the tcpcb for transaction;
777 * set SND.WND = SEG.WND,
778 * initialize CCsend and CCrecv.
780 tp
->t_flags
|= TF_REQ_CC
| TF_RCVD_CC
;
781 tp
->cc_send
= sc
->sc_cc_send
;
782 tp
->cc_recv
= sc
->sc_cc_recv
;
784 if (sc
->sc_flags
& SCF_SACK_PERMITTED
)
785 tp
->t_flags
|= TF_SACK_PERMITTED
;
787 tcp_mss(tp
, sc
->sc_peer_mss
);
790 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment.
792 if (sc
->sc_rxtslot
!= 0)
793 tp
->snd_cwnd
= tp
->t_maxseg
;
794 callout_reset(tp
->tt_keep
, tcp_keepinit
, tcp_timer_keep
, tp
);
796 tcpstat
.tcps_accepts
++;
806 * This function gets called when we receive an ACK for a
807 * socket in the LISTEN state. We look up the connection
808 * in the syncache, and if its there, we pull it out of
809 * the cache and turn it into a full-blown connection in
810 * the SYN-RECEIVED state.
813 syncache_expand(struct in_conninfo
*inc
, struct tcphdr
*th
, struct socket
**sop
,
817 struct syncache_head
*sch
;
820 sc
= syncache_lookup(inc
, &sch
);
823 * There is no syncache entry, so see if this ACK is
824 * a returning syncookie. To do this, first:
825 * A. See if this socket has had a syncache entry dropped in
826 * the past. We don't want to accept a bogus syncookie
827 * if we've never received a SYN.
828 * B. check that the syncookie is valid. If it is, then
829 * cobble up a fake syncache entry, and return.
833 sc
= syncookie_lookup(inc
, th
, *sop
);
837 tcpstat
.tcps_sc_recvcookie
++;
841 * If seg contains an ACK, but not for our SYN/ACK, send a RST.
843 if (th
->th_ack
!= sc
->sc_iss
+ 1)
846 so
= syncache_socket(sc
, *sop
);
850 /* XXXjlemon check this - is this correct? */
851 tcp_respond(NULL
, m
, m
, th
,
852 th
->th_seq
+ tlen
, (tcp_seq
)0, TH_RST
| TH_ACK
);
854 m_freem(m
); /* XXX only needed for above */
855 tcpstat
.tcps_sc_aborted
++;
857 tcpstat
.tcps_sc_completed
++;
862 syncache_drop(sc
, sch
);
868 * Given a LISTEN socket and an inbound SYN request, add
869 * this to the syn cache, and send back a segment:
870 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
873 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
874 * Doing so would require that we hold onto the data and deliver it
875 * to the application. However, if we are the target of a SYN-flood
876 * DoS attack, an attacker could send data which would eventually
877 * consume all available buffer space if it were ACKed. By not ACKing
878 * the data, we avoid this DoS scenario.
881 syncache_add(struct in_conninfo
*inc
, struct tcpopt
*to
, struct tcphdr
*th
,
882 struct socket
**sop
, struct mbuf
*m
)
884 struct tcp_syncache_percpu
*syncache_percpu
;
887 struct syncache
*sc
= NULL
;
888 struct syncache_head
*sch
;
889 struct mbuf
*ipopts
= NULL
;
890 struct rmxp_tao
*taop
;
893 syncache_percpu
= &tcp_syncache_percpu
[mycpu
->gd_cpuid
];
898 * Remember the IP options, if any.
901 if (!inc
->inc_isipv6
)
903 ipopts
= ip_srcroute();
906 * See if we already have an entry for this connection.
907 * If we do, resend the SYN,ACK, and reset the retransmit timer.
910 * The syncache should be re-initialized with the contents
911 * of the new SYN which may have different options.
913 sc
= syncache_lookup(inc
, &sch
);
915 tcpstat
.tcps_sc_dupsyn
++;
918 * If we were remembering a previous source route,
919 * forget it and use the new one we've been given.
922 m_free(sc
->sc_ipopts
);
923 sc
->sc_ipopts
= ipopts
;
926 * Update timestamp if present.
928 if (sc
->sc_flags
& SCF_TIMESTAMP
)
929 sc
->sc_tsrecent
= to
->to_tsval
;
931 /* Just update the TOF_SACK_PERMITTED for now. */
932 if (tcp_do_sack
&& (to
->to_flags
& TOF_SACK_PERMITTED
))
933 sc
->sc_flags
|= SCF_SACK_PERMITTED
;
935 sc
->sc_flags
&= ~SCF_SACK_PERMITTED
;
938 * PCB may have changed, pick up new values.
941 sc
->sc_inp_gencnt
= tp
->t_inpcb
->inp_gencnt
;
942 if (syncache_respond(sc
, m
) == 0) {
943 TAILQ_REMOVE(&syncache_percpu
->timerq
[sc
->sc_rxtslot
],
945 syncache_timeout(syncache_percpu
, sc
, sc
->sc_rxtslot
);
946 tcpstat
.tcps_sndacks
++;
947 tcpstat
.tcps_sndtotal
++;
954 * This allocation is guaranteed to succeed because we
955 * preallocate one more syncache entry than cache_limit.
957 sc
= zalloc(tcp_syncache
.zone
);
960 * Fill in the syncache values.
963 sc
->sc_inp_gencnt
= tp
->t_inpcb
->inp_gencnt
;
964 sc
->sc_ipopts
= ipopts
;
965 sc
->sc_inc
.inc_fport
= inc
->inc_fport
;
966 sc
->sc_inc
.inc_lport
= inc
->inc_lport
;
968 sc
->sc_inc
.inc_isipv6
= inc
->inc_isipv6
;
969 if (inc
->inc_isipv6
) {
970 sc
->sc_inc
.inc6_faddr
= inc
->inc6_faddr
;
971 sc
->sc_inc
.inc6_laddr
= inc
->inc6_laddr
;
972 sc
->sc_route6
.ro_rt
= NULL
;
976 sc
->sc_inc
.inc_faddr
= inc
->inc_faddr
;
977 sc
->sc_inc
.inc_laddr
= inc
->inc_laddr
;
978 sc
->sc_route
.ro_rt
= NULL
;
980 sc
->sc_irs
= th
->th_seq
;
982 sc
->sc_peer_mss
= to
->to_flags
& TOF_MSS
? to
->to_mss
: 0;
984 sc
->sc_iss
= syncookie_generate(sc
);
986 sc
->sc_iss
= karc4random();
988 /* Initial receive window: clip ssb_space to [0 .. TCP_MAXWIN] */
989 win
= ssb_space(&so
->so_rcv
);
991 win
= imin(win
, TCP_MAXWIN
);
994 if (tcp_do_rfc1323
) {
996 * A timestamp received in a SYN makes
997 * it ok to send timestamp requests and replies.
999 if (to
->to_flags
& TOF_TS
) {
1000 sc
->sc_tsrecent
= to
->to_tsval
;
1001 sc
->sc_flags
|= SCF_TIMESTAMP
;
1003 if (to
->to_flags
& TOF_SCALE
) {
1006 /* Compute proper scaling value from buffer space */
1007 while (wscale
< TCP_MAX_WINSHIFT
&&
1008 (TCP_MAXWIN
<< wscale
) < so
->so_rcv
.ssb_hiwat
)
1010 sc
->sc_request_r_scale
= wscale
;
1011 sc
->sc_requested_s_scale
= to
->to_requested_s_scale
;
1012 sc
->sc_flags
|= SCF_WINSCALE
;
1015 if (tcp_do_rfc1644
) {
1017 * A CC or CC.new option received in a SYN makes
1018 * it ok to send CC in subsequent segments.
1020 if (to
->to_flags
& (TOF_CC
| TOF_CCNEW
)) {
1021 sc
->sc_cc_recv
= to
->to_cc
;
1022 sc
->sc_cc_send
= CC_INC(tcp_ccgen
);
1023 sc
->sc_flags
|= SCF_CC
;
1026 if (tcp_do_sack
&& (to
->to_flags
& TOF_SACK_PERMITTED
))
1027 sc
->sc_flags
|= SCF_SACK_PERMITTED
;
1028 if (tp
->t_flags
& TF_NOOPT
)
1029 sc
->sc_flags
= SCF_NOOPT
;
1033 * We have the option here of not doing TAO (even if the segment
1034 * qualifies) and instead fall back to a normal 3WHS via the syncache.
1035 * This allows us to apply synflood protection to TAO-qualifying SYNs
1036 * also. However, there should be a hueristic to determine when to
1037 * do this, and is not present at the moment.
1041 * Perform TAO test on incoming CC (SEG.CC) option, if any.
1042 * - compare SEG.CC against cached CC from the same host, if any.
1043 * - if SEG.CC > chached value, SYN must be new and is accepted
1044 * immediately: save new CC in the cache, mark the socket
1045 * connected, enter ESTABLISHED state, turn on flag to
1046 * send a SYN in the next segment.
1047 * A virtual advertised window is set in rcv_adv to
1048 * initialize SWS prevention. Then enter normal segment
1049 * processing: drop SYN, process data and FIN.
1050 * - otherwise do a normal 3-way handshake.
1052 taop
= tcp_gettaocache(&sc
->sc_inc
);
1053 if (to
->to_flags
& TOF_CC
) {
1054 if ((tp
->t_flags
& TF_NOPUSH
) &&
1055 sc
->sc_flags
& SCF_CC
&&
1056 taop
!= NULL
&& taop
->tao_cc
!= 0 &&
1057 CC_GT(to
->to_cc
, taop
->tao_cc
)) {
1059 so
= syncache_socket(sc
, *sop
);
1061 taop
->tao_cc
= to
->to_cc
;
1065 return (so
!= NULL
);
1069 * No CC option, but maybe CC.NEW: invalidate cached value.
1075 * TAO test failed or there was no CC option,
1076 * do a standard 3-way handshake.
1078 if (syncache_respond(sc
, m
) == 0) {
1079 syncache_insert(sc
, sch
);
1080 tcpstat
.tcps_sndacks
++;
1081 tcpstat
.tcps_sndtotal
++;
1084 tcpstat
.tcps_sc_dropped
++;
1091 syncache_respond(struct syncache
*sc
, struct mbuf
*m
)
1095 u_int16_t tlen
, hlen
, mssopt
;
1096 struct ip
*ip
= NULL
;
1099 struct ip6_hdr
*ip6
= NULL
;
1101 const boolean_t isipv6
= sc
->sc_inc
.inc_isipv6
;
1103 const boolean_t isipv6
= FALSE
;
1107 rt
= tcp_rtlookup6(&sc
->sc_inc
);
1109 mssopt
= rt
->rt_ifp
->if_mtu
-
1110 (sizeof(struct ip6_hdr
) + sizeof(struct tcphdr
));
1112 mssopt
= tcp_v6mssdflt
;
1113 hlen
= sizeof(struct ip6_hdr
);
1115 rt
= tcp_rtlookup(&sc
->sc_inc
);
1117 mssopt
= rt
->rt_ifp
->if_mtu
-
1118 (sizeof(struct ip
) + sizeof(struct tcphdr
));
1120 mssopt
= tcp_mssdflt
;
1121 hlen
= sizeof(struct ip
);
1124 /* Compute the size of the TCP options. */
1125 if (sc
->sc_flags
& SCF_NOOPT
) {
1128 optlen
= TCPOLEN_MAXSEG
+
1129 ((sc
->sc_flags
& SCF_WINSCALE
) ? 4 : 0) +
1130 ((sc
->sc_flags
& SCF_TIMESTAMP
) ? TCPOLEN_TSTAMP_APPA
: 0) +
1131 ((sc
->sc_flags
& SCF_CC
) ? TCPOLEN_CC_APPA
* 2 : 0) +
1132 ((sc
->sc_flags
& SCF_SACK_PERMITTED
) ?
1133 TCPOLEN_SACK_PERMITTED_ALIGNED
: 0);
1135 tlen
= hlen
+ sizeof(struct tcphdr
) + optlen
;
1139 * assume that the entire packet will fit in a header mbuf
1141 KASSERT(max_linkhdr
+ tlen
<= MHLEN
, ("syncache: mbuf too small"));
1144 * XXX shouldn't this reuse the mbuf if possible ?
1145 * Create the IP+TCP header from scratch.
1150 m
= m_gethdr(MB_DONTWAIT
, MT_HEADER
);
1153 m
->m_data
+= max_linkhdr
;
1155 m
->m_pkthdr
.len
= tlen
;
1156 m
->m_pkthdr
.rcvif
= NULL
;
1159 ip6
= mtod(m
, struct ip6_hdr
*);
1160 ip6
->ip6_vfc
= IPV6_VERSION
;
1161 ip6
->ip6_nxt
= IPPROTO_TCP
;
1162 ip6
->ip6_src
= sc
->sc_inc
.inc6_laddr
;
1163 ip6
->ip6_dst
= sc
->sc_inc
.inc6_faddr
;
1164 ip6
->ip6_plen
= htons(tlen
- hlen
);
1165 /* ip6_hlim is set after checksum */
1166 /* ip6_flow = ??? */
1168 th
= (struct tcphdr
*)(ip6
+ 1);
1170 ip
= mtod(m
, struct ip
*);
1171 ip
->ip_v
= IPVERSION
;
1172 ip
->ip_hl
= sizeof(struct ip
) >> 2;
1177 ip
->ip_p
= IPPROTO_TCP
;
1178 ip
->ip_src
= sc
->sc_inc
.inc_laddr
;
1179 ip
->ip_dst
= sc
->sc_inc
.inc_faddr
;
1180 ip
->ip_ttl
= sc
->sc_tp
->t_inpcb
->inp_ip_ttl
; /* XXX */
1181 ip
->ip_tos
= sc
->sc_tp
->t_inpcb
->inp_ip_tos
; /* XXX */
1184 * See if we should do MTU discovery. Route lookups are
1185 * expensive, so we will only unset the DF bit if:
1187 * 1) path_mtu_discovery is disabled
1188 * 2) the SCF_UNREACH flag has been set
1190 if (path_mtu_discovery
1191 && ((sc
->sc_flags
& SCF_UNREACH
) == 0)) {
1192 ip
->ip_off
|= IP_DF
;
1195 th
= (struct tcphdr
*)(ip
+ 1);
1197 th
->th_sport
= sc
->sc_inc
.inc_lport
;
1198 th
->th_dport
= sc
->sc_inc
.inc_fport
;
1200 th
->th_seq
= htonl(sc
->sc_iss
);
1201 th
->th_ack
= htonl(sc
->sc_irs
+ 1);
1202 th
->th_off
= (sizeof(struct tcphdr
) + optlen
) >> 2;
1204 th
->th_flags
= TH_SYN
| TH_ACK
;
1205 th
->th_win
= htons(sc
->sc_wnd
);
1208 /* Tack on the TCP options. */
1211 optp
= (u_int8_t
*)(th
+ 1);
1212 *optp
++ = TCPOPT_MAXSEG
;
1213 *optp
++ = TCPOLEN_MAXSEG
;
1214 *optp
++ = (mssopt
>> 8) & 0xff;
1215 *optp
++ = mssopt
& 0xff;
1217 if (sc
->sc_flags
& SCF_WINSCALE
) {
1218 *((u_int32_t
*)optp
) = htonl(TCPOPT_NOP
<< 24 |
1219 TCPOPT_WINDOW
<< 16 | TCPOLEN_WINDOW
<< 8 |
1220 sc
->sc_request_r_scale
);
1224 if (sc
->sc_flags
& SCF_TIMESTAMP
) {
1225 u_int32_t
*lp
= (u_int32_t
*)(optp
);
1227 /* Form timestamp option as shown in appendix A of RFC 1323. */
1228 *lp
++ = htonl(TCPOPT_TSTAMP_HDR
);
1229 *lp
++ = htonl(ticks
);
1230 *lp
= htonl(sc
->sc_tsrecent
);
1231 optp
+= TCPOLEN_TSTAMP_APPA
;
1235 * Send CC and CC.echo if we received CC from our peer.
1237 if (sc
->sc_flags
& SCF_CC
) {
1238 u_int32_t
*lp
= (u_int32_t
*)(optp
);
1240 *lp
++ = htonl(TCPOPT_CC_HDR(TCPOPT_CC
));
1241 *lp
++ = htonl(sc
->sc_cc_send
);
1242 *lp
++ = htonl(TCPOPT_CC_HDR(TCPOPT_CCECHO
));
1243 *lp
= htonl(sc
->sc_cc_recv
);
1244 optp
+= TCPOLEN_CC_APPA
* 2;
1247 if (sc
->sc_flags
& SCF_SACK_PERMITTED
) {
1248 *((u_int32_t
*)optp
) = htonl(TCPOPT_SACK_PERMITTED_ALIGNED
);
1249 optp
+= TCPOLEN_SACK_PERMITTED_ALIGNED
;
1254 struct route_in6
*ro6
= &sc
->sc_route6
;
1257 th
->th_sum
= in6_cksum(m
, IPPROTO_TCP
, hlen
, tlen
- hlen
);
1258 ip6
->ip6_hlim
= in6_selecthlim(NULL
,
1259 ro6
->ro_rt
? ro6
->ro_rt
->rt_ifp
: NULL
);
1260 error
= ip6_output(m
, NULL
, ro6
, 0, NULL
, NULL
,
1261 sc
->sc_tp
->t_inpcb
);
1263 th
->th_sum
= in_pseudo(ip
->ip_src
.s_addr
, ip
->ip_dst
.s_addr
,
1264 htons(tlen
- hlen
+ IPPROTO_TCP
));
1265 m
->m_pkthdr
.csum_flags
= CSUM_TCP
;
1266 m
->m_pkthdr
.csum_data
= offsetof(struct tcphdr
, th_sum
);
1267 error
= ip_output(m
, sc
->sc_ipopts
, &sc
->sc_route
, 0, NULL
,
1268 sc
->sc_tp
->t_inpcb
);
1276 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .|
1278 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .|
1280 * (A): peer mss index
1284 * The values below are chosen to minimize the size of the tcp_secret
1285 * table, as well as providing roughly a 16 second lifetime for the cookie.
1288 #define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */
1289 #define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */
1291 #define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1)
1292 #define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS)
1293 #define SYNCOOKIE_TIMEOUT \
1294 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT))
1295 #define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK)
1298 u_int32_t ts_secbits
[4];
1300 } tcp_secret
[SYNCOOKIE_NSECRETS
];
1302 static int tcp_msstab
[] = { 0, 536, 1460, 8960 };
1304 static MD5_CTX syn_ctx
;
1306 #define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v))
1309 u_int32_t laddr
, faddr
;
1310 u_int32_t secbits
[4];
1311 u_int16_t lport
, fport
;
1315 CTASSERT(sizeof(struct md5_add
) == 28);
1319 * Consider the problem of a recreated (and retransmitted) cookie. If the
1320 * original SYN was accepted, the connection is established. The second
1321 * SYN is inflight, and if it arrives with an ISN that falls within the
1322 * receive window, the connection is killed.
1324 * However, since cookies have other problems, this may not be worth
1329 syncookie_generate(struct syncache
*sc
)
1331 u_int32_t md5_buffer
[4];
1336 const boolean_t isipv6
= sc
->sc_inc
.inc_isipv6
;
1338 const boolean_t isipv6
= FALSE
;
1341 idx
= ((ticks
<< SYNCOOKIE_TIMESHIFT
) / hz
) & SYNCOOKIE_WNDMASK
;
1342 if (tcp_secret
[idx
].ts_expire
< ticks
) {
1343 for (i
= 0; i
< 4; i
++)
1344 tcp_secret
[idx
].ts_secbits
[i
] = karc4random();
1345 tcp_secret
[idx
].ts_expire
= ticks
+ SYNCOOKIE_TIMEOUT
;
1347 for (data
= sizeof(tcp_msstab
) / sizeof(int) - 1; data
> 0; data
--)
1348 if (tcp_msstab
[data
] <= sc
->sc_peer_mss
)
1350 data
= (data
<< SYNCOOKIE_WNDBITS
) | idx
;
1351 data
^= sc
->sc_irs
; /* peer's iss */
1354 MD5Add(sc
->sc_inc
.inc6_laddr
);
1355 MD5Add(sc
->sc_inc
.inc6_faddr
);
1359 add
.laddr
= sc
->sc_inc
.inc_laddr
.s_addr
;
1360 add
.faddr
= sc
->sc_inc
.inc_faddr
.s_addr
;
1362 add
.lport
= sc
->sc_inc
.inc_lport
;
1363 add
.fport
= sc
->sc_inc
.inc_fport
;
1364 add
.secbits
[0] = tcp_secret
[idx
].ts_secbits
[0];
1365 add
.secbits
[1] = tcp_secret
[idx
].ts_secbits
[1];
1366 add
.secbits
[2] = tcp_secret
[idx
].ts_secbits
[2];
1367 add
.secbits
[3] = tcp_secret
[idx
].ts_secbits
[3];
1369 MD5Final((u_char
*)&md5_buffer
, &syn_ctx
);
1370 data
^= (md5_buffer
[0] & ~SYNCOOKIE_WNDMASK
);
1374 static struct syncache
*
1375 syncookie_lookup(struct in_conninfo
*inc
, struct tcphdr
*th
, struct socket
*so
)
1377 u_int32_t md5_buffer
[4];
1378 struct syncache
*sc
;
1383 data
= (th
->th_ack
- 1) ^ (th
->th_seq
- 1); /* remove ISS */
1384 idx
= data
& SYNCOOKIE_WNDMASK
;
1385 if (tcp_secret
[idx
].ts_expire
< ticks
||
1386 sototcpcb(so
)->ts_recent
+ SYNCOOKIE_TIMEOUT
< ticks
)
1390 if (inc
->inc_isipv6
) {
1391 MD5Add(inc
->inc6_laddr
);
1392 MD5Add(inc
->inc6_faddr
);
1398 add
.laddr
= inc
->inc_laddr
.s_addr
;
1399 add
.faddr
= inc
->inc_faddr
.s_addr
;
1401 add
.lport
= inc
->inc_lport
;
1402 add
.fport
= inc
->inc_fport
;
1403 add
.secbits
[0] = tcp_secret
[idx
].ts_secbits
[0];
1404 add
.secbits
[1] = tcp_secret
[idx
].ts_secbits
[1];
1405 add
.secbits
[2] = tcp_secret
[idx
].ts_secbits
[2];
1406 add
.secbits
[3] = tcp_secret
[idx
].ts_secbits
[3];
1408 MD5Final((u_char
*)&md5_buffer
, &syn_ctx
);
1409 data
^= md5_buffer
[0];
1410 if (data
& ~SYNCOOKIE_DATAMASK
)
1412 data
= data
>> SYNCOOKIE_WNDBITS
;
1415 * This allocation is guaranteed to succeed because we
1416 * preallocate one more syncache entry than cache_limit.
1418 sc
= zalloc(tcp_syncache
.zone
);
1421 * Fill in the syncache values.
1422 * XXX duplicate code from syncache_add
1424 sc
->sc_ipopts
= NULL
;
1425 sc
->sc_inc
.inc_fport
= inc
->inc_fport
;
1426 sc
->sc_inc
.inc_lport
= inc
->inc_lport
;
1428 sc
->sc_inc
.inc_isipv6
= inc
->inc_isipv6
;
1429 if (inc
->inc_isipv6
) {
1430 sc
->sc_inc
.inc6_faddr
= inc
->inc6_faddr
;
1431 sc
->sc_inc
.inc6_laddr
= inc
->inc6_laddr
;
1432 sc
->sc_route6
.ro_rt
= NULL
;
1436 sc
->sc_inc
.inc_faddr
= inc
->inc_faddr
;
1437 sc
->sc_inc
.inc_laddr
= inc
->inc_laddr
;
1438 sc
->sc_route
.ro_rt
= NULL
;
1440 sc
->sc_irs
= th
->th_seq
- 1;
1441 sc
->sc_iss
= th
->th_ack
- 1;
1442 wnd
= ssb_space(&so
->so_rcv
);
1444 wnd
= imin(wnd
, TCP_MAXWIN
);
1448 sc
->sc_peer_mss
= tcp_msstab
[data
];