HardenedBSD/contrib/bind/bin/named/ns_resp.c
Jeroen Ruigrok van der Werven a1f80258eb Virgin import of BIND v8.2.3-T5B
2000-05-26 07:17:19 +00:00

4004 lines
100 KiB
C

#if !defined(lint) && !defined(SABER)
static const char sccsid[] = "@(#)ns_resp.c 4.65 (Berkeley) 3/3/91";
static const char rcsid[] = "$Id: ns_resp.c,v 8.143 2000/05/09 07:38:38 vixie Exp $";
#endif /* not lint */
/*
* Copyright (c) 1986, 1988, 1990
* The Regents of the University of California. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. All advertising materials mentioning features or use of this software
* must display the following acknowledgement:
* This product includes software developed by the University of
* California, Berkeley and its contributors.
* 4. Neither the name of the University nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*/
/*
* Portions Copyright (c) 1993 by Digital Equipment Corporation.
*
* Permission to use, copy, modify, and distribute this software for any
* purpose with or without fee is hereby granted, provided that the above
* copyright notice and this permission notice appear in all copies, and that
* the name of Digital Equipment Corporation not be used in advertising or
* publicity pertaining to distribution of the document or software without
* specific, written prior permission.
*
* THE SOFTWARE IS PROVIDED "AS IS" AND DIGITAL EQUIPMENT CORP. DISCLAIMS ALL
* WARRANTIES WITH REGARD TO THIS SOFTWARE, INCLUDING ALL IMPLIED WARRANTIES
* OF MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL DIGITAL EQUIPMENT
* CORPORATION BE LIABLE FOR ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL
* DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR
* PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS
* ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS
* SOFTWARE.
*/
/*
* Portions Copyright (c) 1995 by International Business Machines, Inc.
*
* International Business Machines, Inc. (hereinafter called IBM) grants
* permission under its copyrights to use, copy, modify, and distribute this
* Software with or without fee, provided that the above copyright notice and
* all paragraphs of this notice appear in all copies, and that the name of IBM
* not be used in connection with the marketing of any product incorporating
* the Software or modifications thereof, without specific, written prior
* permission.
*
* To the extent it has a right to do so, IBM grants an immunity from suit
* under its patents, if any, for the use, sale or manufacture of products to
* the extent that such products are used for performing Domain Name System
* dynamic updates in TCP/IP networks by means of the Software. No immunity is
* granted for any product per se or for any other function of any product.
*
* THE SOFTWARE IS PROVIDED "AS IS", AND IBM DISCLAIMS ALL WARRANTIES,
* INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A
* PARTICULAR PURPOSE. IN NO EVENT SHALL IBM BE LIABLE FOR ANY SPECIAL,
* DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER ARISING
* OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE, EVEN
* IF IBM IS APPRISED OF THE POSSIBILITY OF SUCH DAMAGES.
*/
/*
* Portions Copyright (c) 1996-2000 by Internet Software Consortium.
*
* Permission to use, copy, modify, and distribute this software for any
* purpose with or without fee is hereby granted, provided that the above
* copyright notice and this permission notice appear in all copies.
*
* THE SOFTWARE IS PROVIDED "AS IS" AND INTERNET SOFTWARE CONSORTIUM DISCLAIMS
* ALL WARRANTIES WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES
* OF MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL INTERNET SOFTWARE
* CONSORTIUM BE LIABLE FOR ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL
* DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR
* PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS
* ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS
* SOFTWARE.
*/
#include "port_before.h"
#include <sys/types.h>
#include <sys/param.h>
#include <sys/socket.h>
#include <sys/file.h>
#include <sys/un.h>
#include <netinet/in.h>
#include <arpa/nameser.h>
#include <arpa/inet.h>
#include <errno.h>
#include <limits.h>
#include <resolv.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <syslog.h>
#include <time.h>
#include <isc/eventlib.h>
#include <isc/logging.h>
#include <isc/memcluster.h>
#include <isc/dst.h>
#include "port_after.h"
#include "named.h"
static u_int8_t norootlogged[MAXCLASS]; /* XXX- should be a bitmap */
static const char skipnameFailedAnswer[] = "skipname failed in answer",
skipnameFailedAuth[] = "skipname failed in authority",
skipnameFailedQuery[] = "skipname failed in query",
outofDataQuery[] = "ran out of data in query",
outofDataAnswer[] = "ran out of data in answer",
notSingleQuery[] = "not exactly one query",
expandFailedQuery[] = "dn_expand failed in query",
expandFailedAnswer[] = "dn_expand failed in answer",
expandFailedAuth[] = "dn_expand failed in authority",
outofDataAuth[] = "ran out of data in authority",
dlenOverrunAnswer[] = "dlen overrun in answer",
dlenOverrunAuth[] = "dlen overrun in authority",
dlenUnderrunAnswer[] = "dlen underrun in answer",
outofDataFinal[] = "out of data in final pass",
outofDataAFinal[] = "out of data after final pass",
badNameFound[] = "found an invalid domain name",
wrongQuestion[] = "answer to wrong question",
danglingCname[] = "dangling CNAME pointer",
nonRecursiveForwarder[]= "non-recursive forwarder";
struct db_list {
struct db_list *db_next;
struct databuf *db_dp;
};
struct flush_set {
char * fs_name;
int fs_type;
int fs_class;
u_int fs_cred;
struct db_list *fs_list;
struct db_list *fs_last;
};
static void rrsetadd(struct flush_set *, const char *,
struct databuf *),
rrsetupdate(struct flush_set *, int flags,
struct sockaddr_in, int),
flushrrset(struct flush_set *, struct sockaddr_in),
free_flushset(struct flush_set *, int),
check_hints(struct flush_set *);
static int rrsetcmp(char *, struct db_list *, struct hashbuf *),
check_root(void),
check_ns(void),
wanted(const struct databuf *, int, int),
wantedsig(const struct databuf *, int, int),
rrextract(u_char *, int, u_char *,
struct databuf **, char *, int,
struct sockaddr_in, char **);
static void mark_bad(struct qinfo *qp, struct sockaddr_in from);
static void mark_lame(struct qinfo *qp, struct sockaddr_in from);
static void fast_retry(struct qinfo *qp, struct sockaddr_in from);
static void add_related_additional(char *);
static void free_related_additional(void);
static int related_additional(char *);
static void freestr_maybe(char **);
static enum ordering match_order(const struct namebuf *, int, int);
static int match_name(const struct namebuf *, const char *, size_t);
#define MAX_RELATED 100
static int num_related = 0;
static char *related[MAX_RELATED];
static char *
learntFrom(struct qinfo *qp, struct sockaddr_in *server) {
static char *buf = NULL;
char *a, *ns, *na;
struct databuf *db;
int i;
a = ns = na = "<Not Available>";
for (i = 0; (u_int)i < qp->q_naddr; i++) {
if (ina_equal(qp->q_addr[i].ns_addr.sin_addr,
server->sin_addr)) {
db = qp->q_addr[i].ns;
if (db != NULL) {
if (NS_OPTION_P(OPTION_HOSTSTATS)) {
char nsbuf[20];
if (db->d_ns != NULL) {
strcpy(nsbuf,
inet_ntoa(db->d_ns->addr));
ns = nsbuf;
} else {
ns = zones[db->d_zone]
.z_origin;
}
}
if (db->d_rcode == 0)
na = (char*)qp->q_addr[i].ns->d_data;
}
if (NS_OPTION_P(OPTION_HOSTSTATS)) {
char abuf[20];
db = qp->q_addr[i].nsdata;
if (db != NULL) {
if (db->d_ns != NULL) {
strcpy(abuf,
inet_ntoa(db->d_ns->addr));
a = abuf;
} else {
a = zones[db->d_zone].z_origin;
}
}
}
break;
}
}
if (a == ns && ns == na) /* all "UNKNOWN" */
return (NULL);
if (*a == '\0')
a = "\".\"";
if (*ns == '\0')
ns = "\".\"";
if (*na == '\0')
na = "\".\"";
if (NS_OPTION_P(OPTION_HOSTSTATS)) {
static const char fmt[] = " '%s': learnt (A=%s,NS=%s)";
buf = newstr(sizeof fmt + strlen(na) + strlen(a) + strlen(ns),
0);
if (buf == NULL)
return (NULL);
sprintf(buf, fmt, na, a, ns);
} else {
static const char fmt[] = " '%s'";
buf = newstr(sizeof fmt + strlen(na), 0);
if (buf == NULL)
return (NULL);
sprintf(buf, fmt, na);
}
return (buf);
}
void
ns_resp(u_char *msg, int msglen, struct sockaddr_in from, struct qstream *qsp)
{
struct qinfo *qp;
HEADER *hp;
struct qserv *qs = NULL;
struct databuf *ns, *ns2;
u_char *cp, *answers, *eom = msg + msglen;
struct flush_set *flushset = NULL;
int flushset_size = 0;
struct sockaddr_in *nsa;
struct databuf *nsp[NSMAX];
int i, c, n, qdcount, ancount, aucount, nscount, arcount, arfirst;
int soacount;
u_int qtype, qclass;
int restart; /* flag for processing cname response */
int validanswer, dbflags;
int cname, lastwascname, externalcname;
int count, founddata, foundname;
int buflen;
int newmsglen;
char name[MAXDNAME], qname[MAXDNAME], aname[MAXDNAME];
char msgbuf[MAXDNAME+100];
char *dname, tmpdomain[MAXDNAME];
const char *fname;
const char *formerrmsg = "brain damage";
u_char newmsg[PACKETSZ];
u_char **dpp, *tp;
time_t rtrip;
struct hashbuf *htp;
struct namebuf *np;
struct fwdinfo *fwd;
struct databuf *dp;
int forcecmsg = 0;
char *tname = NULL;
int sendto_errno = 0;
int has_tsig, oldqlen;
u_char *oldqbuf;
u_char *smsg;
int smsglen, smsgsize, siglen;
u_char sig[TSIG_SIG_SIZE];
time_t tsig_time;
DST_KEY *key;
nameserIncr(from.sin_addr, nssRcvdR);
nsp[0] = NULL;
hp = (HEADER *) msg;
if ((qp = qfindid(hp->id)) == NULL ) {
ns_debug(ns_log_default, 1, "DUP? dropped (id %d)",
ntohs(hp->id));
nameserIncr(from.sin_addr, nssRcvdDupR);
return;
}
if (ns_wouldlog(ns_log_default, 2)) {
ns_debug(ns_log_default, 2, "Response (%s %s %s) nsid=%d id=%d",
(qp->q_flags & Q_SYSTEM) ?"SYSTEM" :"USER",
(qp->q_flags & Q_PRIMING) ?"PRIMING" :"NORMAL",
(qp->q_flags & Q_ZSERIAL) ?"ZSERIAL" :"-",
ntohs(qp->q_nsid), ntohs(qp->q_id));
}
if (qp->q_nstsig == NULL)
has_tsig = 0;
else {
int ret;
ret = ns_verify(msg, &msglen, qp->q_nstsig->key,
qp->q_nstsig->sig, qp->q_nstsig->siglen,
NULL, NULL, &tsig_time, 0);
if (ret == 0)
has_tsig = 1;
else {
if (hp->rcode == NOERROR)
hp->rcode = NOTAUTH;
ns_debug(ns_log_default, 1,
"resp: error bad tsig, record dropped");
return;
}
}
/*
* Here we handle high level formatting problems by parsing the header.
*/
qdcount = ntohs(hp->qdcount);
ancount = ntohs(hp->ancount);
aucount = ntohs(hp->nscount);
arcount = ntohs(hp->arcount);
free_addinfo(); /* sets addcount to zero */
cp = msg + HFIXEDSZ;
dpp = dnptrs;
*dpp++ = msg;
if ((*cp & INDIR_MASK) == 0)
*dpp++ = cp;
*dpp = NULL;
if (qdcount == 1) {
n = dn_expand(msg, eom, cp, qname, sizeof(qname));
if (n <= 0) {
formerrmsg = expandFailedQuery;
goto formerr;
}
cp += n;
if (cp + 2 * INT16SZ > eom) {
formerrmsg = outofDataQuery;
goto formerr;
}
GETSHORT(qtype, cp);
GETSHORT(qclass, cp);
if (!ns_nameok(qp, qname, qclass, NULL, response_trans,
ns_ownercontext(qtype, response_trans),
qname, from.sin_addr)) {
formerrmsg = badNameFound;
goto refused;
}
if (cp > eom) {
formerrmsg = outofDataQuery;
goto formerr;
}
if (qp->q_msg && qp->q_msglen &&
!res_nameinquery(qname, qtype, qclass,
qp->q_msg, qp->q_msg + qp->q_msglen)) {
sprintf(msgbuf,
"query section mismatch (%s %s %s)",
qname, p_class(qclass), p_type(qtype));
formerrmsg = msgbuf;
goto formerr;
}
if (ns_samename(qp->q_name, qname) != 1 ||
qp->q_class != qclass ||
qp->q_type != qtype) {
formerrmsg = wrongQuestion;
goto formerr;
}
} else {
strcpy(qname, qp->q_name);
qclass = qp->q_class;
qtype = qp->q_type;
}
/* cp now points after the query section. */
/*
* Here we handle bad responses from servers.
* Several possibilities come to mind:
* The server is sick and returns SERVFAIL
* The server returns some garbage opcode (it's sick)
* The server can't understand our query and return FORMERR
* In all these cases, we drop the packet, disable retries on
* this server and immediately force a retry.
*/
if ((hp->rcode != NOERROR && hp->rcode != NXDOMAIN)
|| (hp->opcode != QUERY
#ifdef BIND_NOTIFY
&& hp->opcode != NS_NOTIFY_OP
#endif
)) {
ns_debug(ns_log_default, 2,
"resp: error (ret %d, op %d), dropped",
hp->rcode, hp->opcode);
switch (hp->rcode) {
case SERVFAIL:
nameserIncr(from.sin_addr, nssRcvdFail);
break;
case FORMERR:
nameserIncr(from.sin_addr, nssRcvdFErr);
break;
default:
nameserIncr(from.sin_addr, nssRcvdErr);
break;
}
if (ns_samename(qp->q_name, qp->q_domain) == 1 &&
hp->rcode == SERVFAIL && hp->opcode == QUERY)
mark_lame(qp, from);
mark_bad(qp, from);
fast_retry(qp, from);
return;
}
if (qdcount != 1) {
/* We don't generate or forward these (yet). */
formerrmsg = notSingleQuery;
goto formerr;
}
/*
* Determine if the response came from a forwarder. Packets from
* anyplace not listed as a forwarder or as a server to whom we
* might have forwarded the query will be dropped.
* XXX - should put this in STATS somewhere.
*/
for (fwd = NS_ZFWDTAB(qp->q_fzone); fwd; fwd = fwd->next)
if (ina_equal(fwd->fwddata->fwdaddr.sin_addr, from.sin_addr))
break;
/*
/*
* find the qinfo pointer and update
* the rtt and fact that we have called on this server before.
*/
{
struct timeval *stp;
for (n = 0, qs = qp->q_addr; (u_int)n < qp->q_naddr; n++, qs++)
if (ina_equal(qs->ns_addr.sin_addr, from.sin_addr))
break;
if ((u_int)n >= qp->q_naddr) {
if (!haveComplained(ina_ulong(from.sin_addr),
(u_long)"unexpected source")) {
ns_info(ns_log_default,
"Response from unexpected source (%s)",
sin_ntoa(from));
}
/*
* We don't know who this response came from so it
* gets dropped on the floor.
*/
return;
}
stp = &qs->stime;
/* Handle response from different (untried) interface. */
if (qs->ns != NULL && stp->tv_sec == 0) {
ns = qs->ns;
while (qs > qp->q_addr
&& (qs->stime.tv_sec == 0 || qs->ns != ns))
qs--;
*stp = qs->stime;
/* XXX - sometimes stp still ends up pointing to
* a zero timeval, in spite of the above attempt.
* Why? What should we do about it?
*/
/* XXX - catch aliases here */
}
/* compute query round trip time */
/* XXX - avoid integer overflow, which is quite likely if stp
* points to a zero timeval (see above).
* rtrip is of type time_t, which we assume is at least
* as big as an int.
*/
if ((tt.tv_sec - stp->tv_sec) > (INT_MAX-999)/1000) {
rtrip = INT_MAX;
} else {
rtrip = ((tt.tv_sec - stp->tv_sec) * 1000 +
(tt.tv_usec - stp->tv_usec) / 1000);
}
if (ns_wouldlog(ns_log_default,3)) {
ns_debug(ns_log_default, 3,
"stime %lu/%lu now %lu/%lu rtt %ld",
(u_long)stp->tv_sec, (u_long)stp->tv_usec,
(u_long)tt.tv_sec, (u_long)tt.tv_usec,
(long)rtrip);
}
/* prevent floating point overflow, limit to 1000 sec */
if (rtrip > 1000000) {
rtrip = 1000000;
}
ns = qs->nsdata;
/*
* Don't update nstime if this doesn't look
* like an address databuf now. XXX
*/
if (ns &&
ns->d_type == T_A &&
ns->d_class == qs->ns->d_class) {
u_long t;
if (ns->d_nstime == 0)
t = rtrip;
else
t = ns->d_nstime * ALPHA
+
(1 - ALPHA) * rtrip;
if (t > 65535)
t = 65535;
ns->d_nstime = (u_int16_t)t;
}
/*
* Record the source so that we do not use this NS again.
*/
if (ns && qs->ns && (qp->q_nusedns < NSMAX)) {
qp->q_usedns[qp->q_nusedns++] = qs->ns;
if (ns_wouldlog(ns_log_default,2)) {
ns_debug(ns_log_default, 2,
"NS #%d addr %s used, rtt %d",
n, sin_ntoa(qs->ns_addr), ns->d_nstime);
}
}
/*
* Penalize those who had earlier chances but failed
* by multiplying round-trip times by BETA (>1).
* Improve nstime for unused addresses by applying GAMMA.
* The GAMMA factor makes unused entries slowly
* improve, so they eventually get tried again.
* GAMMA should be slightly less than 1.
* Watch out for records that may have timed out
* and are no longer the correct type. XXX
*/
for (n = 0, qs = qp->q_addr;
(u_int)n < qp->q_naddr;
n++, qs++) {
u_long t;
ns2 = qs->nsdata;
if (!ns2 || ns2 == ns)
continue;
if (ns2->d_type != T_A ||
ns2->d_class != qs->ns->d_class) /* XXX */
continue;
if (qs->stime.tv_sec) {
if (ns2->d_nstime == 0)
t = (rtrip * BETA);
else
t = ns2->d_nstime * BETA
+
(1 - ALPHA) * rtrip;
} else
t = ns2->d_nstime * GAMMA;
if (t > 65535)
t = 65535;
ns2->d_nstime = (u_int16_t)t;
if (ns_wouldlog(ns_log_default,2)) {
ns_debug(ns_log_default, 2, "NS #%d %s rtt now %d", n,
sin_ntoa(qs->ns_addr),
ns2->d_nstime);
}
}
}
#ifdef BIND_NOTIFY
/*
* For now, NOTIFY isn't defined for ANCOUNT!=0, AUCOUNT!=0,
* or ADCOUNT!=0. Therefore the only real work to be done for
* a NOTIFY-QR is to remove it from the query queue.
*/
if (hp->opcode == NS_NOTIFY_OP) {
ns_info(ns_log_notify,
"Received NOTIFY answer from %s for \"%s %s %s\"",
inet_ntoa(from.sin_addr),
*(qp->q_name) ? qp->q_name : ".",
p_class(qp->q_class), p_type(qp->q_type));
qremove(qp);
return;
}
#endif
if ((qp->q_flags & Q_ZSERIAL) != 0) {
if (hp->aa && ancount > 0 && hp->rcode == NOERROR &&
qtype == T_SOA && (qclass == C_IN || qclass == C_HS))
{
int n;
u_int type, class, dlen;
u_int32_t serial;
u_char *tp = cp;
u_char *rdatap;
n = dn_expand(msg, eom, tp, name, sizeof name);
if (n < 0) {
formerrmsg = expandFailedAnswer;
goto formerr;
}
tp += n; /* name */
if (tp + 3 * INT16SZ + INT32SZ > eom) {
formerrmsg = outofDataAnswer;
goto formerr;
}
GETSHORT(type, tp); /* type */
GETSHORT(class, tp); /* class */
tp += INT32SZ; /* ttl */
GETSHORT(dlen, tp); /* dlen */
rdatap = tp; /* start of rdata */
if (!ns_nameok(qp, name, class, NULL, response_trans,
ns_ownercontext(type, response_trans),
name, from.sin_addr)) {
formerrmsg = badNameFound;
goto refused;
}
if (ns_samename(qname, name) != 1 ||
qtype != type || qclass != class) {
sprintf(msgbuf,
"qserial answer mismatch (%s %s %s)",
name, p_class(class), p_type(type));
formerrmsg = msgbuf;
goto formerr;
}
if (0 >= (n = dn_skipname(tp, eom))) {
formerrmsg = skipnameFailedAnswer;
goto formerr;
}
tp += n; /* mname */
if (0 >= (n = dn_skipname(tp, eom))) {
formerrmsg = skipnameFailedAnswer;
goto formerr;
}
tp += n; /* rname */
if (tp + 5 * INT32SZ > eom) {
formerrmsg = dlenUnderrunAnswer;
goto formerr;
}
GETLONG(serial, tp);
tp += 4 * INT32SZ; /* Skip rest of SOA. */
if ((u_int)(tp - rdatap) != dlen) {
formerrmsg = dlenOverrunAnswer;
goto formerr;
}
for (n = 0, qs = qp->q_addr; (u_int)n < qp->q_naddr;
n++, qs++)
if (ina_equal(qs->ns_addr.sin_addr,
from.sin_addr))
break;
if (n == qp->q_naddr) {
qserial_answer(qp);
qremove(qp);
return;
}
qs->serial = serial;
}
retry(qp);
return;
}
/*
* Non-authoritative, no answer, no error, with referral.
*/
if (hp->rcode == NOERROR && !hp->tc && !hp->aa &&
ancount == 0 && aucount > 0
#ifdef BIND_NOTIFY
&& hp->opcode != NS_NOTIFY_OP
#endif
) {
u_char *tp;
int type, class, dlen;
int foundns, foundsoa;
#ifdef DEBUG
if (debug > 0)
res_pquery(&res, msg, msglen,
log_get_stream(packet_channel));
#endif
/*
* Since there is no answer section (ancount == 0),
* we must be pointing at the authority section (aucount > 0).
*/
tp = cp;
foundns = foundsoa = 0;
for (i = 0 ; i < aucount ; i++) {
n = dn_expand(msg, eom, tp, name, sizeof name);
if (n < 0) {
formerrmsg = expandFailedAuth;
goto formerr;
}
tp += n;
if (tp + 3 * INT16SZ + INT32SZ > eom) {
formerrmsg = outofDataAuth;
goto formerr;
}
GETSHORT(type, tp);
GETSHORT(class, tp);
tp += INT32SZ; /* ttl */
GETSHORT(dlen, tp);
if (!ns_nameok(qp, name, class, NULL, response_trans,
ns_ownercontext(type, response_trans),
name, from.sin_addr)) {
formerrmsg = badNameFound;
goto refused;
}
/* skip rest of record */
if (tp + dlen > eom) {
formerrmsg = outofDataAuth;
goto formerr;
}
tp += dlen;
if (type == T_NS) {
strcpy(aname, name);
foundns = 1;
}
if (type == T_SOA)
foundsoa = 1;
}
/*
* If the answer delegates us either to the same level in
* the hierarchy or closer to the root, we consider this
* server lame. Note that for now we only log the message
* if the T_NS was C_IN, which is technically wrong (NS is
* visible in all classes) but necessary anyway (non-IN
* classes tend to not have good strong delegation graphs).
*/
if (foundns && !foundsoa &&
ns_samedomain(qp->q_domain, aname)) {
if (fwd == NULL) {
nameserIncr(from.sin_addr, nssRcvdLDel);
mark_lame(qp, from);
}
mark_bad(qp, from);
if (class == C_IN && fwd == NULL &&
!haveComplained(ina_ulong(from.sin_addr),
nhash(qp->q_domain))) {
char *learnt_from = learntFrom(qp, &from);
ns_info(ns_log_lame_servers,
"Lame server on '%s' (in '%s'?): %s%s",
qname, qp->q_domain,
sin_ntoa(from),
(learnt_from == NULL) ? "" :
learnt_from);
if (learnt_from != NULL)
freestr(learnt_from);
} else if (fwd != NULL) {
if (!haveComplained(ina_ulong(from.sin_addr),
(u_long)nonRecursiveForwarder))
ns_warning(ns_log_default, "%s: %s",
nonRecursiveForwarder,
sin_ntoa(from));
}
fast_retry(qp, from);
return;
}
}
/*
* Add the info received in the response to the data base.
*/
arfirst = ancount + aucount;
c = arfirst + arcount;
/* Don't return if it's a TSIG signed truncated message */
if (has_tsig > 0 && hp->tc)
goto tcp_retry;
/* -ve $ing non-existence of record, must handle non-authoritative
* NOERRORs with c == 0.
*/
if (!hp->aa && !hp->tc && hp->rcode == NOERROR && c == 0)
goto return_msg;
if (qp->q_flags & Q_SYSTEM)
dbflags = DB_NOTAUTH | DB_NODATA;
else
dbflags = DB_NOTAUTH | DB_NODATA | DB_NOHINTS;
count = c;
if (qp->q_flags & Q_PRIMING)
dbflags |= DB_PRIMING;
if (hp->tc) {
count -= arcount; /* truncation had to affect this */
if (!arcount) {
count -= aucount; /* guess it got this too */
}
if (!(arcount || aucount)) {
count -= ancount; /* things are pretty grim */
}
tcp_retry:
/* retry using tcp provided this was not a tcp query */
if (!(qp->q_flags & Q_USEVC)) {
qp->q_flags |= Q_USEVC;
unsched(qp);
schedretry(qp, 60);
nsa = Q_NEXTADDR(qp, 0);
key = tsig_key_from_addr(nsa->sin_addr);
if (key != NULL) {
smsgsize = qp->q_msglen + TSIG_BUF_SIZE;
smsg = memget(smsgsize);
smsglen = qp->q_msglen;
siglen = sizeof(sig);
memcpy(smsg, qp->q_msg, qp->q_msglen);
n = ns_sign(smsg, &smsglen, smsgsize,
NOERROR, key, NULL, 0,
sig, &siglen, 0);
if (n == 0) {
oldqbuf = qp->q_msg;
oldqlen = qp->q_msglen;
qp->q_msglen = smsglen;
qp->q_msg = smsg;
has_tsig = 1;
qp->q_nstsig = new_tsig(key, sig,
siglen);
}
else {
has_tsig = 0;
free_tsig(qp->q_nstsig);
qp->q_nstsig = NULL;
INSIST(0);
}
}
else {
has_tsig = 0;
free_tsig(qp->q_nstsig);
qp->q_nstsig = NULL;
}
if (tcp_send(qp) != NOERROR)
/*
* We're probably in trouble if tcp_send
* failed, but we'll try to press on because
* there isn't anything else to do.
*/
retry(qp);
if (has_tsig == 1) {
memput(qp->q_msg, smsgsize);
qp->q_msg = oldqbuf;
qp->q_msglen = oldqlen;
}
return;
} else if (!qsp) {
/* outstanding udp response */
return;
}
/* XXX truncated tcp response */
ns_error(ns_log_default,
"ns_resp: TCP truncated: \"%s\" %s %s from %s",
qname, p_class(qclass), p_type(qtype),
sin_ntoa(from));
/* mark this server as bad */
mark_bad(qp, from);
/* try another server, it may have a bigger write buffer */
retry(qp);
return;
}
tp = cp;
restart = 0;
validanswer = 0;
nscount = 0;
soacount = 0;
cname = 0;
lastwascname = 0;
externalcname = 0;
strcpy(aname, qname);
if (count) {
/* allocate 1 extra record for end of set detection */
flushset_size = (count + 1) * sizeof *flushset;
flushset = memget(flushset_size);
if (flushset == NULL)
panic("flushset: out of memory", NULL);
memset(flushset, 0, flushset_size);
} else
flushset = NULL;
for (i = 0; i < count; i++) {
struct databuf *dp;
int type;
freestr_maybe(&tname);
if (cp >= eom) {
free_related_additional();
if (flushset != NULL)
free_flushset(flushset, flushset_size);
formerrmsg = outofDataFinal;
goto formerr;
}
n = rrextract(msg, msglen, cp, &dp, name, sizeof name, from,
&tname);
if (n < 0) {
free_related_additional();
freestr_maybe(&tname);
if (flushset != NULL)
free_flushset(flushset, flushset_size);
formerrmsg = outofDataFinal;
if (hp->rcode == REFUSED)
goto refused;
else
goto formerr;
}
cp += n;
if (!dp)
continue;
type = dp->d_type;
if (i < ancount) {
/* Answer section. */
if (externalcname || ns_samename(name, aname) != 1) {
if (!externalcname)
ns_info(ns_log_resp_checks,
"wrong ans. name (%s != %s)",
name[0] ? name : ".",
aname[0] ? aname : ".");
else
ns_debug(ns_log_resp_checks, 3,
"ignoring answer '%s' after external cname",
name);
db_freedata(dp);
continue;
}
if (type == T_CNAME &&
qtype != T_CNAME && qtype != T_ANY) {
strcpy(aname, (char *)dp->d_data);
if (!ns_samedomain(aname, qp->q_domain))
externalcname = 1;
cname++;
lastwascname = 1;
} else {
validanswer = 1;
lastwascname = 0;
}
if (tname != NULL) {
add_related_additional(tname);
tname = NULL;
}
dp->d_cred = (hp->aa && ns_samename(name, qname) == 1)
? DB_C_AUTH
: DB_C_ANSWER;
} else {
/* After answer section. */
if (lastwascname) {
ns_debug(ns_log_resp_checks, 3,
"last was cname, ignoring auth. and add.");
db_freedata(dp);
break;
}
if (i < arfirst) {
/* Authority section. */
switch (type) {
case T_NS:
case T_SOA:
if (!ns_samedomain(aname, name)) {
ns_info(ns_log_resp_checks,
"bad referral (%s !< %s) from %s",
aname[0] ? aname : ".",
name[0] ? name : ".",
sin_ntoa(from));
db_freedata(dp);
continue;
} else if (fwd == NULL &&
!ns_samedomain(name,
qp->q_domain)) {
if (!externalcname)
ns_info(ns_log_resp_checks,
"bad referral (%s !< %s) from %s",
name[0] ? name : ".",
qp->q_domain[0] ?
qp->q_domain : ".",
sin_ntoa(from));
db_freedata(dp);
continue;
}
if (type == T_NS) {
nscount++;
add_related_additional(tname);
tname = NULL;
}
if (type == T_SOA) {
soacount++;
}
break;
case T_NXT:
/* XXX check */
break;
case T_SIG:
/* XXX check that it relates to an
NS or SOA or NXT */
break;
default:
ns_info(ns_log_resp_checks,
"invalid RR type '%s' in authority section (name = '%s') from %s",
p_type(type), name,
sin_ntoa(from));
db_freedata(dp);
continue;
}
dp->d_cred = (hp->aa && (cname == 0)) ?
DB_C_AUTH : (qp->q_flags & Q_PRIMING)
? DB_C_ANSWER
: DB_C_ADDITIONAL;
} else {
/* Additional section. */
switch (type) {
case T_A:
case T_AAAA:
if (externalcname ||
!ns_samedomain(name, qp->q_domain)) {
ns_debug(ns_log_resp_checks, 3,
"ignoring additional info '%s' type %s",
name, p_type(type));
db_freedata(dp);
continue;
}
if (!related_additional(name)) {
ns_info(ns_log_resp_checks,
"unrelated additional info '%s' type %s from %s",
name, p_type(type),
sin_ntoa(from));
db_freedata(dp);
continue;
}
break;
case T_KEY:
/* XXX check? */
break;
case T_SIG:
/*
* XXX a SIG RR should relate
* to some other RR in this section,
* although if it's the last RR
* it might be a transaction signature.
*/
break;
default:
ns_info(ns_log_resp_checks,
"invalid RR type '%s' in additional section (name = '%s') from %s",
p_type(type), name,
sin_ntoa(from));
db_freedata(dp);
continue;
}
dp->d_cred = (qp->q_flags & Q_PRIMING)
? DB_C_ANSWER
: DB_C_ADDITIONAL;
}
}
rrsetadd(flushset, name, dp);
}
free_related_additional();
freestr_maybe(&tname);
if (flushset != NULL) {
if ((qp->q_flags & Q_SYSTEM) && (qp->q_flags & Q_PRIMING)) {
check_hints(flushset); /* before rrsetupdate */
rrsetupdate(flushset, dbflags, from, 1);
} else
rrsetupdate(flushset, dbflags, from, 0);
free_flushset(flushset, flushset_size);
}
if (lastwascname && !externalcname)
ns_debug(ns_log_cname, 3, "%s (%s) q(%s %s %s) %s qd(%s)",
danglingCname, aname,
(qname && *qname) ? qname : ".",
p_class(qclass), p_type(qtype),
sin_ntoa(from), qp->q_domain);
if (cp > eom) {
formerrmsg = outofDataAFinal;
goto formerr;
}
if ((qp->q_flags & Q_SYSTEM) && ancount) {
if ((qp->q_flags & Q_PRIMING) && !check_root()) {
/* mark server as bad */
mark_bad(qp, from);
fast_retry(qp, from);
return;
}
ns_debug(ns_log_default, 3,
"resp: leaving, SYSQUERY ancount %d", ancount);
#ifdef BIND_NOTIFY
if (qp->q_notifyzone != DB_Z_CACHE) {
struct zoneinfo *zp = &zones[qp->q_notifyzone];
qp->q_notifyzone = DB_Z_CACHE;
ns_notify(zp->z_origin, zp->z_class, ns_t_soa);
}
#endif
qremove(qp);
return;
}
if (ancount && count && !validanswer) {
/*
* Everything passed validation but we didn't get the
* final answer. The response must have contained
* a dangling CNAME. Force a restart of the query.
*
* Don't set restart if count==0, since this means
* the response was truncated in the answer section,
* causing us to set count to 0 which will cause
* validanswer to be 0 as well even though the answer
* section probably contained valid RRs (just not
* a complete set).
* XXX - this works right if we can just forward this
* response to the client, but not if we found a CNAME
* in a prior response and restarted the query.
*/
restart = 1;
}
if (!restart && !qp->q_cmsglen && ancount > 1 && qtype == T_A)
sort_response(tp, eom, ancount, &qp->q_from);
/*
* An answer to a T_ANY query or a successful answer to a
* regular query with no indirection, then just return answer.
*/
if (!restart && ancount && (qtype == T_ANY || !qp->q_cmsglen)) {
ns_debug(ns_log_default, 3,
"resp: got as much answer as there is");
goto return_msg;
}
/*
* We might want to cache this negative answer.
*
* if ancount != 0 and rcode == NOERROR we cannot determine if the
* CNAME chain has been processed to completion or not, so just
* restart the query. DNS needs a NODATA return code!
*
* As some servers incorrectly return a NODATA indication when
* there is a CNAME chain instead of NXDOMAIN, we requery to get
* a definitive answer.
*/
if ((hp->rcode == NXDOMAIN && cname == ancount) ||
(hp->rcode == NOERROR && ancount == 0 &&
(nscount == 0 || soacount != 0)
)
)
{
cache_n_resp(msg, msglen, from, qp->q_name,
qp->q_class, qp->q_type);
if (!qp->q_cmsglen) {
ns_debug(ns_log_default, 3,
"resp: leaving NO: auth = %d", hp->aa);
goto return_msg;
}
forcecmsg = 1;
}
/*
* All messages in here need further processing. i.e. they
* are either CNAMEs or we got referred again.
*/
count = 0;
founddata = 0;
dname = name;
/*
* XXX - the restart stuff doesn't work if any of the answer RRs
* is not cacheable (TTL==0 or unknown RR type), since all of the
* answer must pass through the cache and be re-assembled.
*/
if (qp->q_cmsglen != 0) {
ns_debug(ns_log_default, 1, "Cname second pass");
newmsglen = MIN(PACKETSZ, qp->q_cmsglen);
memcpy(newmsg, qp->q_cmsg, newmsglen);
} else {
newmsglen = MIN(PACKETSZ, msglen);
memcpy(newmsg, msg, newmsglen);
}
hp = (HEADER *) newmsg;
hp->ancount = htons(0);
hp->nscount = htons(0);
hp->arcount = htons(0);
hp->rcode = NOERROR;
dnptrs[0] = newmsg;
dnptrs[1] = NULL;
cp = newmsg + HFIXEDSZ;
/*
* Keep in mind that none of this code works when QDCOUNT>1.
* cp ends up pointed just past the query section in both cases.
*/
/*
* Arrange for dname to contain the query name. The query
* name can be either the original query name if restart==0
* or the target of the last CNAME if we are following a
* CNAME chain and were referred.
*/
n = dn_expand(newmsg, newmsg + newmsglen, cp, dname, sizeof name);
if (n < 0) {
ns_debug(ns_log_default, 1, "dn_expand failed");
goto servfail;
}
if (!res_dnok(dname)) {
ns_debug(ns_log_default, 1, "bad name (%s)", dname);
goto servfail;
}
cp += n + QFIXEDSZ;
buflen = sizeof(newmsg) - (cp - newmsg);
cname = 0;
try_again:
ns_debug(ns_log_default, 1, "resp: nlookup(%s) qtype=%d", dname,
qtype);
foundname = 0;
fname = "";
htp = hashtab; /* lookup relative to root */
np = nlookup(dname, &htp, &fname, 0);
ns_debug(ns_log_default, 1, "resp: %s '%s' as '%s' (cname=%d)",
np == NULL ? "missed" : "found", dname, fname, cname);
if (np == NULL || fname != dname)
goto fetch_ns;
foundname++;
answers = cp;
count = cp - newmsg;
/*
* Look for NXDOMAIN record.
*/
for (dp = np->n_data; dp; dp = dp->d_next) {
if (!stale(dp) && (dp->d_rcode == NXDOMAIN) &&
(dp->d_class == (int)qclass)) {
#ifdef RETURNSOA
n = finddata(np, qclass, T_SOA, hp, &dname,
&buflen, &count);
if ( n != 0) {
if (count) {
cp += n;
buflen -= n;
newmsglen += n;
hp->nscount = htons((u_int16_t)count);
}
if (hp->rcode == NOERROR_NODATA) {
hp->rcode = NOERROR;
goto return_newmsg;
}
}
#else
count = 0;
#endif
hp->rcode = NXDOMAIN;
/*
* XXX forcing AA all the time isn't right, but
* we have to work that way by default
* for compatibility with older servers.
*/
if (!NS_OPTION_P(OPTION_NONAUTH_NXDOMAIN))
hp->aa = 1;
ns_debug(ns_log_default, 3, "resp: NXDOMAIN aa = %d",
hp->aa);
if ((count == 0) || NS_OPTION_P(OPTION_NORFC2308_TYPE1))
goto return_newmsg;
founddata = 1;
goto fetch_ns;
}
}
n = finddata(np, qclass, qtype, hp, &dname, &buflen, &count);
if (n == 0)
goto fetch_ns; /* NO data available */
if (hp->rcode) {
if (hp->rcode == NOERROR_NODATA)
hp->rcode = NOERROR;
#ifdef RETURNSOA
if (count) {
cp += n;
buflen -= n;
hp->nscount = htons((u_int16_t)count);
}
#endif
if ((count == 0) || NS_OPTION_P(OPTION_NORFC2308_TYPE1))
goto return_newmsg;
founddata = 1;
goto fetch_ns;
}
cp += n;
buflen -= n;
hp->ancount = htons(ntohs(hp->ancount) + (u_int16_t)count);
if (fname != dname && qtype != T_CNAME && qtype != T_ANY) {
cname++;
goto try_again;
}
founddata = 1;
ns_debug(ns_log_default, 3,
"resp: foundname=%d, count=%d, founddata=%d, cname=%d",
foundname, count, founddata, cname);
if (count > 1 && qtype == T_A)
sort_response(answers, cp, count, &qp->q_from);
fetch_ns:
if (hp->tc)
goto return_newmsg;
/*
* Look for name servers to refer to and fill in the authority
* section or record the address for forwarding the query
* (recursion desired).
*/
free_nsp(nsp);
switch (findns(&np, qclass, nsp, &count, 0)) {
case NXDOMAIN: /* shouldn't happen */
ns_debug(ns_log_default, 3, "req: leaving (%s, rcode %d)",
dname, hp->rcode);
if (!foundname)
hp->rcode = NXDOMAIN;
if (qclass != C_ANY) {
hp->aa = 1;
if (np && (!foundname || !founddata)) {
n = doaddauth(hp, cp, buflen, np, nsp[0]);
cp += n;
buflen -= n;
}
}
goto return_newmsg;
case SERVFAIL:
goto servfail;
}
if (founddata) {
hp = (HEADER *)newmsg;
n = add_data(np, nsp, cp, buflen, &count);
if (n < 0) {
hp->tc = 1;
n = (-n);
}
cp += n;
buflen -= n;
hp->nscount = htons((u_int16_t)count + ntohs(hp->nscount));
goto return_newmsg;
}
/*
* If we get here, we don't have the answer yet and are about
* to iterate to try and get it. First, infinite loop avoidance.
*/
if (qp->q_nqueries++ > MAXQUERIES) {
ns_debug(ns_log_default, 1,
"resp: MAXQUERIES exceeded (%s %s %s)",
dname, p_class(qclass), p_type(qtype));
ns_info(ns_log_default,
"MAXQUERIES exceeded, possible data loop in resolving (%s)",
dname);
goto servfail;
}
/* Reset the query control structure */
ns_freeqns(qp, "ns_resp");
qp->q_naddr = 0;
qp->q_curaddr = 0;
nsfwdadd(qp, NS_ZFWDTAB(qp->q_fzone));
if (qp->q_domain != NULL)
freestr(qp->q_domain);
getname(np, tmpdomain, sizeof tmpdomain);
qp->q_domain = savestr(tmpdomain, 1);
if (NS_ZOPTION_P(qp->q_fzone, OPTION_FORWARD_ONLY))
n = 0;
else if ((n = nslookup(nsp, qp, dname, "ns_resp")) <= 0) {
if (n < 0) {
if (n == -1)
ns_debug(ns_log_default, 3,
"resp: nslookup reports danger");
if (cname) /* a remote CNAME that does not have data */
goto return_newmsg;
goto servfail;
} else {
ns_debug(ns_log_default, 3,
"resp: no addrs found for NS's");
/*
* Timeout while sysquery looks up the NS addresses.
*
* Hopefully we'll have them when the client asks
* again.
*
* too bad we can't just wait for the sysquery
* response to restart this query (it's too hard).
*
* We could try to crawl back up the tree looking
* for reachable servers, but we may have just
* gotten delegated down here by a response with
* no A RRs for the servers. If we blindly tried
* this strategy, we bang on the same server forever.
*/
goto timeout;
}
}
for (n = 0; (u_int)n < qp->q_naddr; n++)
qp->q_addr[n].stime.tv_sec = 0;
qp->q_addr[0].stime = tt;
if (cname) {
if (qp->q_cname++ == MAXCNAMES) {
ns_debug(ns_log_default, 3,
"resp: leaving, MAXCNAMES exceeded");
goto servfail;
}
ns_debug(ns_log_default, 1, "q_cname = %d", qp->q_cname);
ns_debug(ns_log_default, 3,
"resp: building recursive query; nslookup");
if (qp->q_cmsg == NULL) {
qp->q_cmsg = qp->q_msg;
qp->q_cmsglen = qp->q_msglen;
qp->q_cmsgsize = qp->q_msgsize;
} else if (qp->q_msg != NULL)
memput(qp->q_msg, qp->q_msgsize);
qp->q_msg = (u_char *)memget(PACKETSZ);
if (qp->q_msg == NULL) {
ns_notice(ns_log_default, "resp: memget error");
goto servfail;
}
qp->q_msgsize = PACKETSZ;
n = res_nmkquery(&res, QUERY, dname, qclass, qtype,
NULL, 0, NULL, qp->q_msg, PACKETSZ);
if (n < 0) {
ns_info(ns_log_default, "resp: res_mkquery(%s) failed",
dname);
goto servfail;
}
if (qp->q_name != NULL)
freestr(qp->q_name);
qp->q_name = savestr(dname, 1);
qp->q_msglen = n;
hp = (HEADER *) qp->q_msg;
hp->rd = 0;
} else
hp = (HEADER *) qp->q_msg;
hp->id = qp->q_nsid = htons(nsid_next());
hp->rd = (qp->q_addr[0].forwarder ? 1 : 0);
unsched(qp);
schedretry(qp, retrytime(qp));
nsa = Q_NEXTADDR(qp, 0);
if (ns_wouldlog(ns_log_default,1)) {
ns_debug(ns_log_default, 1,
"resp: forw -> %s ds=%d nsid=%d id=%d %dms",
sin_ntoa(*nsa), ds,
ntohs(qp->q_nsid), ntohs(qp->q_id),
(qp->q_addr[0].nsdata != NULL)
? qp->q_addr[0].nsdata->d_nstime
: -1);
}
#ifdef DEBUG
if (debug >= 10)
res_pquery(&res, qp->q_msg, qp->q_msglen,
log_get_stream(packet_channel));
#endif
key = tsig_key_from_addr(nsa->sin_addr);
if (key != NULL) {
smsgsize = qp->q_msglen + TSIG_BUF_SIZE;
smsg = memget(smsgsize);
smsglen = qp->q_msglen;
siglen = sizeof(sig);
memcpy(smsg, qp->q_msg, qp->q_msglen);
n = ns_sign(smsg, &smsglen, smsgsize, NOERROR, key, NULL, 0,
sig, &siglen, 0);
if (n == 0) {
oldqbuf = qp->q_msg;
oldqlen = qp->q_msglen;
qp->q_msglen = smsglen;
qp->q_msg = smsg;
has_tsig = 1;
qp->q_nstsig = new_tsig(key, sig, siglen);
}
else {
has_tsig = 0;
free_tsig(qp->q_nstsig);
qp->q_nstsig = NULL;
INSIST(0);
}
}
else {
has_tsig = 0;
free_tsig(qp->q_nstsig);
qp->q_nstsig = NULL;
}
if (qp->q_flags & Q_USEVC) {
if (tcp_send(qp) != NOERROR) {
if (!haveComplained(ina_ulong(nsa->sin_addr),
(u_long)tcpsendStr))
ns_info(ns_log_default,
"ns_forw: tcp_send(%s) failed: %s",
sin_ntoa(*nsa), strerror(errno));
}
} else if (sendto(ds, (char*)qp->q_msg, qp->q_msglen, 0,
(struct sockaddr *)nsa,
sizeof(struct sockaddr_in)) < 0)
{
sendto_errno = errno;
if (!haveComplained(ina_ulong(nsa->sin_addr),
(u_long)sendtoStr))
ns_info(ns_log_default, "ns_resp: sendto(%s): %s",
sin_ntoa(*nsa), strerror(errno));
nameserIncr(nsa->sin_addr, nssSendtoErr);
}
if (has_tsig == 1) {
memput(qp->q_msg, smsgsize);
qp->q_msg = oldqbuf;
qp->q_msglen = oldqlen;
}
hp->rd = 0; /* leave set to 0 for dup detection */
nameserIncr(nsa->sin_addr, nssSentFwdR);
nameserIncr(qp->q_from.sin_addr, nssRcvdFwdR);
ns_debug(ns_log_default, 3, "resp: Query sent.");
free_nsp(nsp);
switch (sendto_errno) {
case ENETDOWN:
case ENETUNREACH:
case EHOSTDOWN:
case EHOSTUNREACH:
unsched(qp);
schedretry(qp, (time_t) 0);
}
return;
formerr:
if (!haveComplained(ina_ulong(from.sin_addr), (u_long)formerrmsg))
ns_info(ns_log_resp_checks, "Malformed response from %s (%s)",
sin_ntoa(from), formerrmsg);
fast_retry(qp, from);
free_nsp(nsp);
return;
return_msg:
nameserIncr(from.sin_addr, nssRcvdFwdR);
nameserIncr(qp->q_from.sin_addr, nssSentFwdR);
nameserIncr(qp->q_from.sin_addr, nssSentAns);
if (!hp->aa)
nameserIncr(qp->q_from.sin_addr, nssSentNaAns);
if (hp->rcode == NXDOMAIN)
nameserIncr(qp->q_from.sin_addr, nssSentNXD);
/* The "standard" return code */
hp->qr = 1;
hp->id = qp->q_id;
hp->rd = 1;
hp->ra = (NS_OPTION_P(OPTION_NORECURSE) == 0);
(void) send_msg(msg, msglen, qp);
qremove(qp);
free_nsp(nsp);
return;
return_newmsg:
nameserIncr(qp->q_from.sin_addr, nssSentAns);
if (!hp->aa)
nameserIncr(qp->q_from.sin_addr, nssSentNaAns);
if (hp->rcode == NXDOMAIN)
nameserIncr(qp->q_from.sin_addr, nssSentNXD);
n = doaddinfo(hp, cp, buflen);
cp += n;
buflen -= n;
hp->qr = 1;
hp->id = qp->q_id;
hp->rd = 1;
hp->ra = (NS_OPTION_P(OPTION_NORECURSE) == 0);
(void) send_msg(newmsg, cp - newmsg, qp);
qremove(qp);
free_nsp(nsp);
return;
refused:
hp = (HEADER *)(qp->q_cmsglen ? qp->q_cmsg : qp->q_msg);
hp->rcode = REFUSED;
hp->qr = 1;
hp->id = qp->q_id;
hp->rd = 1;
hp->ra = (NS_OPTION_P(OPTION_NORECURSE) == 0);
(void) send_msg((u_char *)hp,
(qp->q_cmsglen ? qp->q_cmsglen : qp->q_msglen),
qp);
qremove(qp);
free_nsp(nsp);
return;
servfail:
nameserIncr(qp->q_from.sin_addr, nssSentFail);
hp = (HEADER *)(qp->q_cmsglen ? qp->q_cmsg : qp->q_msg);
hp->rcode = SERVFAIL;
hp->qr = 1;
hp->id = qp->q_id;
hp->rd = 1;
hp->ra = (NS_OPTION_P(OPTION_NORECURSE) == 0);
(void) send_msg((u_char *)hp,
(qp->q_cmsglen ? qp->q_cmsglen : qp->q_msglen),
qp);
qremove(qp);
free_nsp(nsp);
return;
timeout:
if (qp->q_stream)
sq_remove(qp->q_stream);
qremove(qp);
free_nsp(nsp);
return;
}
#define BOUNDS_CHECK(ptr, count) \
do { \
if ((ptr) + (count) > eom) { \
hp->rcode = FORMERR; \
return (-1); \
} \
} while (0)
static int
rrextract(u_char *msg, int msglen, u_char *rrp, struct databuf **dpp,
char *dname, int namelen, struct sockaddr_in from, char **tnamep)
{
u_char *cp, *eom, *rdatap;
u_int class, type, dlen;
int n, n1, n2;
u_int32_t ttl;
u_char *cp1, data[MAXDATA*2];
HEADER *hp = (HEADER *)msg;
enum context context;
if (tnamep != NULL)
*tnamep = NULL;
*dpp = NULL;
cp = rrp;
eom = msg + msglen;
if ((n = dn_expand(msg, eom, cp, dname, namelen)) < 0) {
hp->rcode = FORMERR;
return (-1);
}
cp += n;
BOUNDS_CHECK(cp, 2*INT16SZ + INT32SZ + INT16SZ);
GETSHORT(type, cp);
GETSHORT(class, cp);
if (class > CLASS_MAX) {
ns_debug(ns_log_default, 3, "bad class in rrextract");
hp->rcode = FORMERR;
return (-1);
}
GETLONG(ttl, cp);
if (ttl > MAXIMUM_TTL) {
ns_debug(ns_log_default, 5, "%s: converted TTL > %u to 0",
dname, MAXIMUM_TTL);
ttl = 0;
}
GETSHORT(dlen, cp);
BOUNDS_CHECK(cp, dlen);
rdatap = cp;
if (!ns_nameok(NULL, dname, class, NULL, response_trans,
ns_ownercontext(type, response_trans),
dname, from.sin_addr)) {
hp->rcode = REFUSED;
return (-1);
}
ns_debug(ns_log_default, 3,
"rrextract: dname %s type %d class %d ttl %d",
dname, type, class, ttl);
/*
* Convert the resource record data into the internal
* database format.
*
* On entry to the switch:
* CP points to the RDATA section of the wire-format RR.
* DLEN is its length.
* The memory area at DATA is available for processing.
*
* On exit from the switch:
* CP has been incremented past the RR.
* CP1 points to the RDATA section of the database-format RR.
* N contains the length of the RDATA section of the dbase-format RR.
*
* The new data at CP1 for length N will be copied into the database,
* so it need not be in any particular storage location.
*/
switch (type) {
case T_A:
if (dlen != INT32SZ) {
hp->rcode = FORMERR;
return (-1);
}
/*FALLTHROUGH*/
case T_WKS:
case T_HINFO:
case T_TXT:
case T_X25:
case T_ISDN:
case T_NSAP:
case T_AAAA:
case T_LOC:
case T_KEY:
case ns_t_cert:
cp1 = cp;
n = dlen;
cp += n;
break;
case T_CNAME:
case T_MB:
case T_MG:
case T_MR:
case T_NS:
case T_PTR:
n = dn_expand(msg, eom, cp, (char *)data, sizeof data);
if (n < 0) {
hp->rcode = FORMERR;
return (-1);
}
if (!ns_nameok(NULL, (char *)data, class, NULL, response_trans,
type == T_PTR ?ns_ptrcontext(dname) :domain_ctx,
dname, from.sin_addr)) {
hp->rcode = FORMERR;
return (-1);
}
cp += n;
cp1 = data;
n = strlen((char *)data) + 1;
if (tnamep != NULL && (type == T_NS || type == T_MB))
*tnamep = savestr((char *)cp1, 1);
break;
case T_SOA:
context = hostname_ctx;
goto soa_rp_minfo;
case T_RP:
case T_MINFO:
context = mailname_ctx;
/* FALLTHROUGH */
soa_rp_minfo:
n = dn_expand(msg, eom, cp, (char *)data, sizeof data);
if (n < 0) {
hp->rcode = FORMERR;
return (-1);
}
if (!ns_nameok(NULL, (char *)data, class, NULL, response_trans,
context, dname, from.sin_addr)) {
hp->rcode = FORMERR;
return (-1);
}
cp += n;
/*
* The next use of 'cp' is dn_expand(), so we don't have
* to BOUNDS_CHECK() here.
*/
cp1 = data + (n = strlen((char *)data) + 1);
n1 = sizeof(data) - n;
if (type == T_SOA)
n1 -= 5 * INT32SZ;
n = dn_expand(msg, eom, cp, (char *)cp1, n1);
if (n < 0) {
hp->rcode = FORMERR;
return (-1);
}
if (type == T_RP)
context = domain_ctx;
else
context = mailname_ctx;
if (!ns_nameok(NULL, (char *)cp1, class, NULL, response_trans,
context, dname, from.sin_addr)) {
hp->rcode = FORMERR;
return (-1);
}
cp += n;
cp1 += strlen((char *)cp1) + 1;
if (type == T_SOA) {
n = 5 * INT32SZ;
BOUNDS_CHECK(cp, n);
memcpy(cp1, cp, n);
cp += n;
cp1 += n;
}
n = cp1 - data;
cp1 = data;
break;
case T_NAPTR:
/* Grab weight and port. */
BOUNDS_CHECK(cp, INT16SZ*2);
memcpy(data, cp, INT16SZ*2);
cp1 = data + INT16SZ*2;
cp += INT16SZ*2;
/* Flags */
BOUNDS_CHECK(cp, 1);
n = *cp++;
BOUNDS_CHECK(cp, n);
*cp1++ = n;
memcpy(cp1, cp, n);
cp += n; cp1 += n;
/* Service */
BOUNDS_CHECK(cp, 1);
n = *cp++;
BOUNDS_CHECK(cp, n);
*cp1++ = n;
memcpy(cp1, cp, n);
cp += n; cp1 += n;
/* Regexp */
BOUNDS_CHECK(cp, 1);
n = *cp++;
BOUNDS_CHECK(cp, n);
*cp1++ = n;
memcpy(cp1, cp, n);
cp += n; cp1 += n;
/* Replacement */
n = dn_expand(msg, eom, cp, (char *)cp1,
sizeof data - (cp1 - data));
if (n < 0) {
hp->rcode = FORMERR;
return (-1);
}
if (!ns_nameok(NULL, (char *)cp1, class, NULL, response_trans,
hostname_ctx, dname, from.sin_addr)) {
hp->rcode = FORMERR;
return (-1);
}
cp += n;
/* compute end of data */
cp1 += strlen((char *)cp1) + 1;
/* compute size of data */
n = cp1 - data;
cp1 = data;
break;
case T_MX:
case T_AFSDB:
case T_RT:
case T_SRV:
/* grab preference */
BOUNDS_CHECK(cp, INT16SZ);
memcpy(data, cp, INT16SZ);
cp1 = data + INT16SZ;
cp += INT16SZ;
if (type == T_SRV) {
/* Grab weight and port. */
BOUNDS_CHECK(cp, INT16SZ*2);
memcpy(cp1, cp, INT16SZ*2);
cp1 += INT16SZ*2;
cp += INT16SZ*2;
}
/* get name */
n = dn_expand(msg, eom, cp, (char *)cp1,
sizeof data - (cp1 - data));
if (n < 0) {
hp->rcode = FORMERR;
return (-1);
}
if (!ns_nameok(NULL, (char *)cp1, class, NULL, response_trans,
hostname_ctx, dname, from.sin_addr)) {
hp->rcode = FORMERR;
return (-1);
}
cp += n;
if (tnamep != NULL)
*tnamep = savestr((char *)cp1, 1);
/* compute end of data */
cp1 += strlen((char *)cp1) + 1;
/* compute size of data */
n = cp1 - data;
cp1 = data;
break;
case T_PX:
/* grab preference */
BOUNDS_CHECK(cp, INT16SZ);
memcpy(data, cp, INT16SZ);
cp1 = data + INT16SZ;
cp += INT16SZ;
/* get MAP822 name */
n = dn_expand(msg, eom, cp, (char *)cp1,
sizeof data - INT16SZ);
if (n < 0) {
hp->rcode = FORMERR;
return (-1);
}
if (!ns_nameok(NULL, (char *)cp1, class, NULL, response_trans,
domain_ctx, dname, from.sin_addr)) {
hp->rcode = FORMERR;
return (-1);
}
cp += n;
/*
* The next use of 'cp' is dn_expand(), so we don't have
* to BOUNDS_CHECK() here.
*/
cp1 += (n = strlen((char *)cp1) + 1);
n1 = sizeof(data) - n;
n = dn_expand(msg, eom, cp, (char *)cp1, n1);
if (n < 0) {
hp->rcode = FORMERR;
return (-1);
}
if (!ns_nameok(NULL, (char *)cp1, class, NULL, response_trans,
domain_ctx, dname, from.sin_addr)) {
hp->rcode = FORMERR;
return (-1);
}
cp += n;
cp1 += strlen((char *)cp1) + 1;
n = cp1 - data;
cp1 = data;
break;
case T_SIG: {
u_long origTTL, exptime, signtime, timetilexp, now;
u_int8_t alg;
/* Check signature time, expiration, and adjust TTL. */
/* This code is similar to that in db_load.c. */
/* Skip coveredType, save alg, skip labels */
BOUNDS_CHECK(cp, INT16SZ + 1 + 1 + 3*INT32SZ);
cp1 = cp + INT16SZ;
alg = *cp1++;
cp1++;
GETLONG(origTTL, cp1);
GETLONG(exptime, cp1);
GETLONG(signtime, cp1);
now = time(NULL); /* Get current time in GMT/UTC */
/* Don't let bogus name servers increase the signed TTL */
if (ttl > origTTL) {
ns_debug(ns_log_default, 3,
"shrinking SIG TTL from %d to origTTL %d",
ttl, origTTL);
ttl = origTTL;
}
/* Don't let bogus signers "sign" in the future. */
if (signtime > now) {
ns_debug(ns_log_default, 3,
"ignoring SIG: signature date %s is in the future",
p_secstodate (signtime));
return ((cp - rrp) + dlen);
}
/* Ignore received SIG RR's that are already expired. */
if (exptime <= now) {
ns_debug(ns_log_default, 3,
"ignoring SIG: expiration %s is in the past",
p_secstodate (exptime));
return ((cp - rrp) + dlen);
}
/* Lop off the TTL at the expiration time. */
timetilexp = exptime - now;
if (timetilexp < ttl) {
ns_debug(ns_log_default, 3,
"shrinking expiring %s SIG TTL from %d to %d",
p_secstodate (exptime), ttl, timetilexp);
ttl = timetilexp;
}
/* The following code is copied from named-xfer.c. */
cp1 = (u_char *)data;
/* first just copy over the type_covered, algorithm, */
/* labels, orig ttl, two timestamps, and the footprint */
BOUNDS_CHECK(cp, 18);
memcpy(cp1, cp, 18);
cp += 18;
cp1 += 18;
/* then the signer's name */
n = dn_expand(msg, eom, cp, (char *)cp1, (sizeof data) - 18);
if (n < 0 || n + NS_SIG_SIGNER > dlen) {
hp->rcode = FORMERR;
return (-1);
}
cp += n;
cp1 += strlen((char*)cp1)+1;
/* finally, we copy over the variable-length signature.
Its size is the total data length, minus what we copied. */
n = dlen - (NS_SIG_SIGNER + n);
if (n > (sizeof data) - (cp1 - (u_char *)data)) {
hp->rcode = FORMERR;
return (-1); /* out of room! */
}
switch (alg) {
case NS_ALG_MD5RSA:
if (n < NS_MD5RSA_MIN_SIZE || n > NS_MD5RSA_MAX_SIZE)
hp->rcode = FORMERR;
break;
case NS_ALG_DSA:
if (n != NS_DSA_SIG_SIZE)
hp->rcode = FORMERR;
break;
default:
break;
}
if (hp->rcode == FORMERR)
return (-1);
memcpy(cp1, cp, n);
cp += n;
cp1 += n;
/* compute size of data */
n = cp1 - (u_char *)data;
cp1 = (u_char *)data;
break;
}
case T_NXT:
n = dn_expand(msg, eom, cp, (char *)data, sizeof data);
/*
* By testing if n >= dlen, we are requiring that the type
* bitmap be at least one octet. This is reasonable
* because we always have to look at the 0 bit to see if
* this is a "different format" NXT or not.
*/
if (n < 0 || n >= dlen) {
hp->rcode = FORMERR;
return (-1);
}
if (!ns_nameok(NULL, (char *)data, class, NULL, response_trans,
domain_ctx, dname, from.sin_addr)) {
hp->rcode = FORMERR;
return (-1);
}
cp += n;
n1 = strlen((char *)data) + 1;
cp1 = data + n1;
/*
* We don't need to BOUNDS_CHECK() cp here because we've
* previously checked that 'dlen' bytes are in bounds, and
* we know that n < dlen.
*/
n2 = dlen - n;
/*
* The first bit of the first octet determines the format
* of the NXT record. A format for types >= 128 has not
* yet been defined, so if bit zero is set, we just copy
* what's there because we don't understand it.
*/
if ((*cp & 0x80) == 0) {
/*
* Bit zero is not set; this is an ordinary NXT
* record. The bitmap must be at least 4 octets
* because the NXT bit should be set. It should be
* less than or equal to 16 octets because this NXT
* format is only defined for types < 128.
*/
if (n2 < 4 || n2 > 16) {
hp->rcode = FORMERR;
return (-1);
}
}
if (n2 > sizeof data - n1) {
hp->rcode = FORMERR;
return (-1);
}
memcpy(cp1, cp, n2);
cp += n2;
cp1 += n2;
/* compute size of data */
n = cp1 - (u_char *)data;
cp1 = (u_char *)data;
break;
default:
ns_debug(ns_log_default, 3, "unknown type %d", type);
return ((cp - rrp) + dlen);
}
if (cp > eom) {
hp->rcode = FORMERR;
return (-1);
}
if ((u_int)(cp - rdatap) != dlen) {
ns_debug(ns_log_default, 3,
"encoded rdata length is %u, but actual length was %u",
dlen, (u_int)(cp - rdatap));
hp->rcode = FORMERR;
return (-1);
}
if (n > MAXDATA) {
ns_debug(ns_log_default, 1,
"update type %d: %d bytes is too much data",
type, n);
hp->rcode = FORMERR;
return (-1);
}
ttl += tt.tv_sec;
*dpp = savedata(class, type, ttl, cp1, n);
return (cp - rrp);
}
int
send_msg(u_char *msg, int msglen, struct qinfo *qp) {
HEADER *hp = (HEADER *) msg;
u_char *oldmsg;
int oldlen;
int msgsize;
int ret;
if (qp->q_flags & Q_SYSTEM)
return (1);
if (!qp->q_stream && (msglen > PACKETSZ))
msglen = trunc_adjust(msg, msglen, PACKETSZ);
if (ns_wouldlog(ns_log_default, 1)) {
ns_debug(ns_log_default, 1, "send_msg -> %s (%s %d) id=%d",
sin_ntoa(qp->q_from),
qp->q_stream == NULL ? "UDP" : "TCP",
qp->q_stream == NULL ? qp->q_dfd : qp->q_stream->s_rfd,
ntohs(qp->q_id));
}
#ifdef DEBUG
if (ns_wouldlog(ns_log_default, 4)) {
struct qinfo *tqp;
for (tqp = nsqhead; tqp != NULL; tqp = tqp->q_link) {
ns_debug(ns_log_default, 4,
"qp %#lx q_id: %d q_nsid: %d q_msglen: %d",
(u_long)tqp, tqp->q_id,
tqp->q_nsid, tqp->q_msglen);
ns_debug(ns_log_default, 4,
"\tq_naddr: %d q_curaddr: %d",
tqp->q_naddr, tqp->q_curaddr);
ns_debug(ns_log_default, 4,
"\tq_next: %#lx q_link: %#lx",
(u_long)qp->q_next, (u_long)qp->q_link);
}
}
if (debug >= 6)
res_pquery(&res, msg, msglen, log_get_stream(packet_channel));
#endif /* DEBUG */
if (qp->q_tsig != NULL) {
u_char sig[TSIG_SIG_SIZE];
int siglen = sizeof(sig);
oldmsg = msg;
oldlen = msglen;
msgsize = msglen + TSIG_BUF_SIZE;
msg = memget(msgsize);
memcpy(msg, oldmsg, oldlen);
ret = ns_sign(msg, &msglen, msgsize, NOERROR, qp->q_tsig->key,
qp->q_tsig->sig, qp->q_tsig->siglen,
sig, &siglen, 0);
if (ret != 0) {
INSIST(0);
}
}
if (qp->q_stream == NULL) {
/*
* Don't send FORMERR to these well known ports
* (loop avoidance).
*/
switch (ntohs(qp->q_from.sin_port)) {
case 7: /* echo */
case 13: /* daytime */
case 19: /* chargen */
case 37: /* time */
if (hp->rcode == FORMERR)
return (-1);
default:
break;
}
if (sendto(qp->q_dfd, (char*)msg, msglen, 0,
(struct sockaddr *)&qp->q_from,
sizeof(qp->q_from)) < 0) {
if (!haveComplained(ina_ulong(qp->q_from.sin_addr),
(u_long)sendtoStr))
#if defined(SPURIOUS_ECONNREFUSED)
if (errno != ECONNREFUSED)
#endif
ns_info(ns_log_default,
"send_msg: sendto(%s): %s",
sin_ntoa(qp->q_from),
strerror(errno));
nameserIncr(qp->q_from.sin_addr, nssSendtoErr);
return (1);
}
} else
writestream(qp->q_stream, (u_char*)msg, msglen);
if (qp->q_tsig != NULL)
memput(msg, oldlen + TSIG_BUF_SIZE);
return (0);
}
static int
root_server_p(ns_class class) {
struct zoneinfo *zp = find_zone("", class);
return (zp != NULL &&
(zp->z_type == z_master || zp->z_type == z_slave));
}
void
prime_cache(void) {
int root = root_server_p(ns_c_in);
ns_debug(ns_log_default, 1, "prime_cache: priming = %d, root = %d",
priming, root);
if (!priming && !root) {
struct qinfo *qp = sysquery("", ns_c_in, ns_t_ns,
NULL, 0, ns_port, ns_o_query);
if (qp != NULL) {
qp->q_flags |= (Q_SYSTEM | Q_PRIMING);
priming++;
}
}
needs_prime_cache = 0;
}
struct qinfo *
sysquery(const char *dname, int class, int type,
struct in_addr *nss, int nsc, u_int16_t port, int opcode)
{
struct qinfo *qp, *oqp;
HEADER *hp;
char tmpdomain[MAXDNAME];
struct namebuf *np = NULL;
struct databuf *nsp[NSMAX];
struct hashbuf *htp1;
struct hashbuf *htp2;
struct hashbuf *htp3;
struct sockaddr_in *nsa;
const char *fname;
int n, count;
int sendto_errno = 0;
u_char *oldqbuf;
int oldqlen, has_tsig;
u_char *smsg;
int smsglen, smsgsize, siglen;
u_char sig[TSIG_SIG_SIZE];
DST_KEY *key;
nsp[0] = NULL;
ns_debug(ns_log_default, 3, "sysquery(%s, %d, %d, %#x, %d, %d)",
dname, class, type, nss, nsc, ntohs(port));
qp = qnew(dname, class, type, (nss != NULL && nsc != 0) ? 0 : 1);
if (nss != NULL && nsc != 0)
np = NULL;
else if (!NS_ZOPTION_P(qp->q_fzone, OPTION_FORWARD_ONLY)) {
htp1 = hashtab;
htp2 = hashtab;
htp3 = fcachetab;
if (priming && dname[0] == '\0') {
np = NULL;
} else if (((np = nlookup(dname, &htp1, &fname, 0)) == NULL) &&
((np = nlookup("", &htp2, &fname, 0)) == NULL) &&
((np = nlookup("", &htp3, &fname, 0)) == NULL)) {
ns_info(ns_log_default,
"sysquery: nlookup error on %s?",
dname);
err1:
ns_freeqry(qp);
return (NULL);
}
n = findns(&np, class, nsp, &count, 0);
switch (n) {
case NXDOMAIN:
case SERVFAIL:
ns_info(ns_log_default,
"sysquery: findns error (%s) on %s?",
n == NXDOMAIN ? "NXDOMAIN" : "SERVFAIL",
dname);
err2:
free_nsp(nsp);
goto err1;
}
}
/* Build new qinfo struct. */
qp->q_cmsg = qp->q_msg = NULL;
qp->q_dfd = ds;
if (nss == NULL || nsc == 0)
nsfwdadd(qp, NS_ZFWDTAB(qp->q_fzone));
qp->q_expire = tt.tv_sec + RETRY_TIMEOUT*2;
qp->q_flags |= Q_SYSTEM;
getname(np, tmpdomain, sizeof tmpdomain);
qp->q_domain = savestr(tmpdomain, 1);
if ((qp->q_msg = (u_char *)memget(PACKETSZ)) == NULL) {
ns_notice(ns_log_default, "sysquery: memget failed");
goto err2;
}
qp->q_msgsize = PACKETSZ;
n = res_nmkquery(&res, opcode, dname, class,
type, NULL, 0, NULL,
qp->q_msg, PACKETSZ);
if (n < 0) {
ns_info(ns_log_default,
"sysquery: res_mkquery(%s) failed", dname);
goto err2;
}
qp->q_msglen = n;
hp = (HEADER *) qp->q_msg;
hp->id = qp->q_nsid = htons(nsid_next());
hp->rd = (qp->q_addr[qp->q_curaddr].forwarder ? 1 : 0);
/* First check for an already pending query for this data. */
for (oqp = nsqhead; oqp != NULL; oqp = oqp->q_link) {
if ((oqp != qp)
&& (oqp->q_msglen == qp->q_msglen)
&& memcmp(oqp->q_msg+2, qp->q_msg + 2,
qp->q_msglen - 2) == 0
) {
#ifdef BIND_NOTIFY
/* XXX - need fancier test to suppress duplicate
* NOTIFYs to the same server (compare nss?)
*/
if (opcode != NS_NOTIFY_OP)
#endif /*BIND_NOTIFY*/
{
ns_debug(ns_log_default, 3,
"sysquery: duplicate");
goto err2;
}
}
}
if (nss != NULL && nsc != 0) {
int i;
struct qserv *qs;
for (i = 0, qs = qp->q_addr; i < nsc; i++, qs++) {
qs->ns_addr.sin_family = AF_INET;
qs->ns_addr.sin_addr = nss[i];
qs->ns_addr.sin_port = port;
qs->ns = NULL;
qs->nsdata = NULL;
qs->stime = tt;
qs->forwarder = 0;
qs->nretry = 0;
}
qp->q_naddr = nsc;
} else if (!NS_ZOPTION_P(qp->q_fzone, OPTION_FORWARD_ONLY)) {
fetch_a:
count = nslookup(nsp, qp, dname, "sysquery");
if (count <= 0) {
if (count < 0) {
if (n == -1)
ns_info(ns_log_default,
"sysquery: nslookup reports danger (%s)",
dname);
goto err2;
} else if (np && NAME(*np)[0] == '\0') {
/*
* It's not too serious if we don't have
* the root server addresses if we have to
* go through a forwarder anyway. Don't
* bother to log it, since prime_cache()
* won't do anything about it as currently
* implemented.
*
* XXX - should we skip setting
* needs_prime_cache as well?
*
* XXX - what happens when we implement
* selective forwarding?
*/
if (!NS_OPTION_P(OPTION_FORWARD_ONLY))
ns_warning(ns_log_default,
"sysquery: no addrs found for root NS (%s)",
dname);
if (class == C_IN && !priming)
needs_prime_cache = 1;
goto err2;
}
if (np) {
free_nsp(nsp);
nsp[0] = NULL;
np = np_parent(np);
n = findns(&np, class, nsp, &count, 0);
switch (n) {
case NXDOMAIN: /*FALLTHROUGH*/
case SERVFAIL:
ns_info(ns_log_default,
"sysquery: findns error (%d) on %s?",
n, dname);
goto err2;
}
getname(np, tmpdomain, sizeof tmpdomain);
if (qp->q_domain != NULL)
freestr(qp->q_domain);
qp->q_domain = savestr(tmpdomain, 1);
goto fetch_a;
}
goto err2;
}
}
schedretry(qp, retrytime(qp));
qp->q_addr[0].stime = tt; /* XXX - why not every? */
nsa = Q_NEXTADDR(qp, 0);
ns_debug(ns_log_default, 1,
"sysquery: send -> %s dfd=%d nsid=%d id=%d retry=%ld",
sin_ntoa(*nsa), qp->q_dfd,
ntohs(qp->q_nsid), ntohs(qp->q_id),
(long)qp->q_time);
#ifdef DEBUG
if (debug >= 10)
res_pquery(&res, qp->q_msg, qp->q_msglen,
log_get_stream(packet_channel));
#endif
key = tsig_key_from_addr(nsa->sin_addr);
if (key != NULL) {
smsgsize = qp->q_msglen + TSIG_BUF_SIZE;
smsg = memget(smsgsize);
smsglen = qp->q_msglen;
siglen = sizeof(sig);
memcpy(smsg, qp->q_msg, qp->q_msglen);
n = ns_sign(smsg, &smsglen, smsgsize, NOERROR, key, NULL, 0,
sig, &siglen, 0);
if (n == 0) {
oldqbuf = qp->q_msg;
oldqlen = qp->q_msglen;
qp->q_msglen = smsglen;
qp->q_msg = smsg;
has_tsig = 1;
qp->q_nstsig = new_tsig(key, sig, siglen); /* BEW? */
}
else {
INSIST(0);
has_tsig = 0;
free_tsig(qp->q_nstsig);
qp->q_nstsig = NULL;
}
}
else {
has_tsig = 0;
free_tsig(qp->q_nstsig);
qp->q_nstsig = NULL;
}
if (sendto(qp->q_dfd, (char*)qp->q_msg, qp->q_msglen, 0,
(struct sockaddr *)nsa,
sizeof(struct sockaddr_in)) < 0) {
sendto_errno = errno;
if (!haveComplained(ina_ulong(nsa->sin_addr),
(u_long)sendtoStr))
ns_info(ns_log_default, "sysquery: sendto(%s): %s",
sin_ntoa(*nsa), strerror(errno));
nameserIncr(nsa->sin_addr, nssSendtoErr);
}
if (has_tsig == 1) {
memput(qp->q_msg, smsgsize);
qp->q_msg = oldqbuf;
qp->q_msglen = oldqlen;
}
nameserIncr(nsa->sin_addr, nssSentSysQ);
free_nsp(nsp);
switch (sendto_errno) {
case ENETDOWN:
case ENETUNREACH:
case EHOSTDOWN:
case EHOSTUNREACH:
unsched(qp);
schedretry(qp, (time_t) 0);
}
return (qp);
}
/*
* Check the list of root servers after receiving a response
* to a query for the root servers.
*/
static int
check_root() {
struct databuf *dp, *pdp;
struct namebuf *np;
int count = 0;
priming = 0;
for (np = hashtab->h_tab[0]; np != NULL; np = np->n_next)
if (NAME(*np)[0] == '\0')
break;
if (np == NULL) {
ns_notice(ns_log_default, "check_root: Can't find root!");
return (0);
}
for (dp = np->n_data; dp != NULL; dp = dp->d_next)
if (dp->d_type == T_NS)
count++;
ns_debug(ns_log_default, 1, "%d root servers", count);
if (count < server_options->minroots) {
ns_notice(ns_log_default,
"check_root: %d root servers after query to root server < min",
count);
return (0);
}
pdp = NULL;
dp = np->n_data;
while (dp != NULL) {
if (dp->d_type == T_NS && dp->d_zone == DB_Z_CACHE &&
dp->d_ttl < (u_int32_t)tt.tv_sec) {
ns_debug(ns_log_default, 1,
"deleting old root server '%s'",
dp->d_data);
dp = rm_datum(dp, np, pdp, NULL);
/* SHOULD DELETE FROM HINTS ALSO */
continue;
}
pdp = dp;
dp = dp->d_next;
}
if (check_ns())
return (1);
else {
priming = 1;
return (0);
}
}
/*
* Check the root to make sure that for each NS record we have a A RR
*/
static int
check_ns() {
struct databuf *dp, *tdp;
struct namebuf *np, *tnp;
struct hashbuf *htp;
char *dname;
int found_arr;
const char *fname;
time_t curtime;
int servers = 0, rrsets = 0;
ns_debug(ns_log_default, 2, "check_ns()");
curtime = (u_int32_t) tt.tv_sec;
for (np = hashtab->h_tab[0]; np != NULL; np = np->n_next) {
if (NAME(*np)[0] != '\0')
continue;
for (dp = np->n_data; dp != NULL; dp = dp->d_next) {
int cnames = 0;
if (dp->d_rcode)
continue;
if (dp->d_type != T_NS)
continue;
servers++;
/* look for A records */
dname = (caddr_t) dp->d_data;
htp = hashtab;
tnp = nlookup(dname, &htp, &fname, 0);
if (tnp == NULL || fname != dname) {
ns_debug(ns_log_default, 3,
"check_ns: %s: not found %s %#lx",
dname, fname, (u_long)tnp);
sysquery(dname, dp->d_class, T_A, NULL,
0, ns_port, QUERY);
continue;
}
/* look for name server addresses */
found_arr = 0;
(void)delete_stale(tnp);
for (tdp = tnp->n_data;
tdp != NULL;
tdp = tdp->d_next) {
if (tdp->d_rcode)
continue;
if (tdp->d_type == T_CNAME)
cnames++;
if (tdp->d_type != T_A ||
tdp->d_class != dp->d_class)
continue;
if ((tdp->d_zone == DB_Z_CACHE) &&
(tdp->d_ttl < (u_int32_t)curtime)) {
ns_debug(ns_log_default, 3,
"check_ns: stale entry '%s'",
NAME(*tnp));
found_arr = 0;
break;
}
found_arr++;
}
if (found_arr)
rrsets++;
else if (cnames > 0)
ns_info(ns_log_default,
"Root NS %s -> CNAME %s",
NAME(*np), NAME(*tnp));
else
sysquery(dname, dp->d_class, T_A, NULL,
0, ns_port, QUERY);
}
}
ns_debug(ns_log_default, 2, "check_ns: %d %d", servers, rrsets);
return ((servers <= 2)
? (rrsets == servers)
: ((rrsets * 2) >= servers)
);
}
/* int findns(npp, class, nsp, countp, flag)
* Find NS's or an SOA
* npp, class:
* dname whose most enclosing NS is wanted
* nsp, countp:
* result array and count; array will also be NULL terminated
* flag:
* boolean: we're being called from ADDAUTH, bypass authority checks
* return value:
* NXDOMAIN: we are authoritative for this {dname,class}
* *countp is bogus, but nsp[] has a single SOA returned in it.
* SERVFAIL: we are auth but zone isn't loaded; or, no root servers found
* *countp and nsp[] are bogus.
* OK: we are not authoritative, and here are the NS records we found.
* *countp and nsp[] return NS records of interest.
*/
int
findns(struct namebuf **npp, int class,
struct databuf **nsp, int *countp, int flag)
{
struct namebuf *np = *npp;
struct databuf *dp;
struct databuf **nspp;
struct hashbuf *htp;
nsp[0] = NULL;
if (priming && (np == NULL || NAME(*np)[0] == '\0'))
htp = fcachetab;
else
htp = hashtab;
try_again:
if (htp == fcachetab && class == C_IN && !priming)
/*
* XXX - do we want to set needs_prime_cache if
* OPTION_FORWARD_ONLY?
*/
needs_prime_cache = 1;
if (np == NULL) {
/* find the root */
for (np = htp->h_tab[0]; np != NULL; np = np->n_next)
if (NAME(*np)[0] == '\0')
break;
}
while (np != NULL) {
ns_debug(ns_log_default, 5, "findns: np %#x '%s'", np,
NAME(*np));
/* Look first for SOA records. */
#ifdef ADDAUTH
if (!flag)
#endif
for (dp = np->n_data; dp != NULL; dp = dp->d_next) {
if (dp->d_zone != DB_Z_CACHE &&
((zones[dp->d_zone].z_type == Z_PRIMARY) ||
(zones[dp->d_zone].z_type == Z_SECONDARY)) &&
match(dp, class, T_SOA) && dp->d_type == T_SOA) {
ns_debug(ns_log_default, 3,
"findns: SOA found");
if (zones[dp->d_zone].z_flags & Z_AUTH) {
*npp = np;
nsp[0] = dp;
nsp[1] = NULL;
DRCNTINC(dp);
return (NXDOMAIN);
} else {
/* XXX: zone isn't loaded but we're
* primary or secondary for it.
* should we fwd this?
*/
return (SERVFAIL);
}
}
}
/* If no SOA records, look for NS records. */
nspp = &nsp[0];
*nspp = NULL;
(void)delete_stale(np);
for (dp = np->n_data; dp != NULL; dp = dp->d_next) {
if (!match(dp, class, T_NS))
continue;
if (dp->d_rcode)
continue;
/*
* Don't use records that may become invalid to
* reference later when we do the rtt computation.
* Never delete our safety-belt information!
*
* XXX: this is horribly bogus.
*/
if ((dp->d_zone == DB_Z_CACHE) &&
(dp->d_ttl < (u_int32_t)tt.tv_sec) &&
!(dp->d_flags & DB_F_HINT)) {
ns_debug(ns_log_default, 1,
"findns: stale entry '%s'",
NAME(*np));
/*
* We may have already added NS databufs
* and are going to throw them away. Fix
* reference counts. We don't need to free
* them here as we just got them from the
* cache.
*/
while (nspp > &nsp[0]) {
nspp--;
DRCNTDEC(*nspp);
}
nsp[0] = NULL;
goto try_parent;
}
if (nspp < &nsp[NSMAX-1]) {
*nspp++ = dp;
DRCNTINC(dp);
}
}
*countp = nspp - nsp;
if (*countp > 0) {
ns_debug(ns_log_default, 3,
"findns: %d NS's added for '%s'",
*countp, NAME(*np));
*nspp = NULL;
*npp = np;
return (OK); /* Success, got some NS's */
}
try_parent:
np = np_parent(np);
}
if (htp == hashtab) {
htp = fcachetab;
goto try_again;
}
ns_debug(ns_log_default, 1,
"findns: No root nameservers for class %s?", p_class(class));
if ((unsigned)class < MAXCLASS && norootlogged[class] == 0) {
norootlogged[class] = 1;
ns_info(ns_log_default, "No root nameservers for class %s",
p_class(class));
}
return (SERVFAIL);
}
/*
* Extract RR's from the given node that match class and type.
* Return number of bytes added to response.
* If no matching data is found, then 0 is returned.
*/
int
finddata(struct namebuf *np, int class, int type,
HEADER *hp, char **dnamep, int *lenp, int *countp)
{
struct databuf *dp;
char *cp;
int buflen, n, count = 0;
char *new_dnamep = NULL;
int defer = 0, found_count = 0, choice, i;
struct databuf **found = NULL;
struct databuf **tmpfound = NULL;
int foundcname;
int stalecount;
int ret = 0;
stalecount = delete_stale(np);
/* We don't want to return cached SIG records when asked for SIGs,
* since we may have an incomplete set.
*/
if (type == T_SIG && findMyZone(np, class) == DB_Z_CACHE)
return(0);
if (type != T_ANY && type != T_PTR && type != T_NXT) {
found = memget((stalecount + 1) * sizeof *found);
tmpfound = memget((stalecount + 1) * sizeof *tmpfound);
if (found == NULL || tmpfound == NULL)
ns_panic(ns_log_default, 1, "finddata: out of memory");
defer = 1;
}
buflen = *lenp;
#ifdef DEBUG
if (buflen > PACKETSZ)
ns_debug(ns_log_default, 1, "finddata(): buflen=%d", buflen);
#endif
cp = ((char *)hp) + *countp;
foundcname = 0;
for (dp = np->n_data; dp != NULL; dp = dp->d_next) {
if (!wanted(dp, class, type)) {
if (type == T_CNAME && class == dp->d_class) {
/* any data means no CNAME exists */
if (dp->d_type != T_NXT &&
dp->d_type != T_KEY &&
dp->d_type != T_SIG) {
ret = 0;
goto done;
}
}
continue;
}
if (dp->d_cred == DB_C_ADDITIONAL) {
#ifdef NOADDITIONAL
continue;
#else
/* we want to expire additional data very
* quickly. current strategy is to cut 5%
* off each time it is accessed. this makes
* stale(dp) true earlier when this datum is
* used often.
*/
dp->d_ttl = tt.tv_sec
+
0.95 * (int) (dp->d_ttl - tt.tv_sec);
#endif
}
/* -ve $ing stuff, anant@isi.edu
* if we have a -ve $ed record, change the rcode on the
* header to reflect that
*/
if (dp->d_rcode == NOERROR_NODATA) {
if (count != 0) {
/*
* This should not happen, yet it does...
*/
ns_info(ns_log_default,
"NODATA & data for \"%s\" type %d class %d",
*dnamep, type, class);
continue;
}
if (type == T_ANY)
continue;
hp->rcode = NOERROR_NODATA;
if (dp->d_size == 0) { /* !RETURNSOA */
ret = 1;
goto done;
}
}
if (dp->d_rcode == NXDOMAIN) {
if (count != 0) {
/*
* This should not happen, yet it might...
*/
ns_info(ns_log_default,
"NXDOMAIN & data for \"%s\" type %d class %d",
*dnamep, type, class);
continue;
}
hp->rcode = NXDOMAIN;
if (dp->d_size == 0) { /* !RETURNSOA */
ret = 1;
goto done;
}
}
/* Don't put anything but key or sig RR's in response to
requests for key or sig */
if (((type == T_SIG) || (type == T_KEY)) &&
(!((dp->d_type == T_SIG) || (dp->d_type == T_KEY))) )
continue;
if (!defer) {
if (foundcname != 0 && dp->d_type == T_CNAME)
continue;
if ((n = make_rr(*dnamep, dp, (u_char *)cp, buflen, 1,
dnptrs, dnptrs_end, 0)) < 0) {
hp->tc = 1;
ret = *lenp - buflen;
goto done;
}
if (dp->d_secure != DB_S_SECURE)
hp->ad = 0;
cp += n;
buflen -= n;
count++;
if (dp->d_type == T_CNAME) {
foundcname = 1;
#define FOLLOWCNAME(type) \
(type != T_KEY) && (type != T_SIG) && (type != T_NXT) && (type != T_ANY)
/* don't alias if querying for key, sig, nxt, or any */
if (FOLLOWCNAME(type))
new_dnamep = (char *)dp->d_data;
}
} else {
if (dp->d_type == T_CNAME)
foundcname = 1;
found[found_count++] = dp;
}
}
if (found_count == 0 && count == 0) {
ret = 0;
goto done;
}
/*
* If the query type was SIG or ANY we will have returned the SIG
* records already.
*/
if (type != T_SIG && type != T_ANY) {
for (dp = np->n_data; dp != NULL; dp = dp->d_next) {
if (!wantedsig(dp, class, type))
continue;
if (dp->d_cred == DB_C_ADDITIONAL) {
#ifdef NOADDITIONAL
continue;
#else
/* we want to expire additional data very
* quickly. current strategy is to cut 5%
* off each time it is accessed. this makes
* stale(dp) true earlier when this datum is
* used often.
*/
dp->d_ttl = tt.tv_sec
+
0.95 * (int) (dp->d_ttl - tt.tv_sec);
#endif
}
if (!defer) {
if ((n = make_rr(*dnamep, dp, (u_char *)cp,
buflen, 1, dnptrs, dnptrs_end,
0)) < 0) {
hp->tc = 1;
ret = *lenp - buflen;
goto done;
}
if (dp->d_secure != DB_S_SECURE)
hp->ad = 0;
cp += n;
buflen -= n;
count++;
} else
found[found_count++] = dp;
}
}
if (defer && found_count > 0) {
int first_sig;
int non_sig_count;
int sig_count; /* number of SIG records in found */
int idx, jdx;
enum ordering order;
order = match_order(np, class, foundcname ? T_CNAME : type);
/* shuffle the SIG records down to the bottom of the array
* as we need to make sure they get packed last, no matter
* what the ordering is. We're sure to maintain the
* original ordering within the two sets of records (so
* that fixed_order can work).
* First we pack the non-SIG records into the temp array.
*/
for (idx = jdx = 0 ; idx < found_count ; idx++) {
if (found[idx]->d_type != T_SIG) {
tmpfound[jdx++] = found[idx];
}
}
non_sig_count = jdx;
sig_count = found_count - jdx;
first_sig = jdx ;
/* now shift the SIG records down to the end of the array
* and copy in the non-SIG records
*/
for (i = idx = found_count - 1 ; idx >= 0 ; idx--) {
if (i < non_sig_count) {
found[i] = tmpfound[i];
i--;
} else if (found[idx]->d_type == T_SIG) {
found[i--] = found[idx] ;
}
}
foundcname = 0;
switch (order) {
case fixed_order:
for (i = 0; i < found_count; i++) {
dp = found[i];
if (foundcname != 0 && dp->d_type == T_CNAME)
continue;
if (dp->d_type == T_CNAME) {
foundcname = 1;
if (FOLLOWCNAME(type)) {
new_dnamep = (char *)dp->d_data;
}
}
if ((n = make_rr(*dnamep, dp, (u_char *)cp,
buflen, 1,
dnptrs, dnptrs_end, 0)) < 0) {
hp->tc = 1;
ret = *lenp - buflen;
goto done;
}
if (dp->d_secure != DB_S_SECURE)
hp->ad = 0;
cp += n;
buflen -= n;
count++;
}
break;
case random_order: {
/* first we shuffle the non-SIG records */
int iters = non_sig_count;
for (i = 0; i < iters; i++) {
choice = ((u_int)rand()>>3) % non_sig_count;
non_sig_count--;
dp = found[choice];
found[choice] = found[non_sig_count];
if (foundcname != 0 && dp->d_type == T_CNAME)
continue;
if (dp->d_type == T_CNAME) {
foundcname = 1;
if (FOLLOWCNAME(type)) {
new_dnamep = (char *)dp->d_data;
}
}
if ((n = make_rr(*dnamep, dp, (u_char *)cp,
buflen, 1,
dnptrs, dnptrs_end, 0)) < 0) {
hp->tc = 1;
ret = *lenp - buflen;
goto done;
}
if (dp->d_secure != DB_S_SECURE)
hp->ad = 0;
cp += n;
buflen -= n;
count++;
}
/* now shuffle the SIG records */
iters = sig_count;
for (i = 0; i < iters; i++) {
choice = ((u_int)rand()>>3) % sig_count;
choice += first_sig;
sig_count--;
dp = found[choice];
found[choice] = found[sig_count + first_sig];
if ((n = make_rr(*dnamep, dp, (u_char *)cp,
buflen, 1,
dnptrs, dnptrs_end, 0)) < 0) {
hp->tc = 1;
ret = *lenp - buflen;
goto done;
}
if (dp->d_secure != DB_S_SECURE)
hp->ad = 0;
cp += n;
buflen -= n;
count++;
}
break;
}
case cyclic_order:
/* first we do the non-SIG records */
if (non_sig_count > 0)
choice = ((u_int)rand()>>3) % non_sig_count;
else
choice = 0;
for (i = 0; i < non_sig_count ; i++) {
dp = found[(i + choice) % non_sig_count];
if (foundcname != 0 && dp->d_type == T_CNAME)
continue;
if (dp->d_type == T_CNAME) {
foundcname = 1;
if (FOLLOWCNAME(type)) {
new_dnamep = (char *)dp->d_data;
}
}
if ((n = make_rr(*dnamep, dp, (u_char *)cp,
buflen, 1,
dnptrs, dnptrs_end, 0)) < 0) {
hp->tc = 1;
ret = *lenp - buflen;
goto done;
}
if (dp->d_secure != DB_S_SECURE)
hp->ad = 0;
cp += n;
buflen -= n;
count++;
}
/* now do the SIG record rotation. */
if (sig_count > 0) {
choice = ((u_int)rand()>>3) % sig_count;
choice += first_sig;
i = choice;
do {
dp = found[i];
if ((n = make_rr(*dnamep, dp,
(u_char *)cp,
buflen, 1,
dnptrs,
dnptrs_end, 0)) < 0) {
hp->tc = 1;
ret = *lenp - buflen;
goto done;
}
if (dp->d_secure != DB_S_SECURE)
hp->ad = 0;
cp += n;
buflen -= n;
count++;
i++;
if (i >= found_count)
i = first_sig;
} while (i != choice);
}
break;
default:
ns_warning(ns_log_default, "finddata: unknown ordering: %d",
order);
break;
}
}
if (new_dnamep != NULL)
*dnamep = new_dnamep;
ns_debug(ns_log_default, 3, "finddata: added %d class %d type %d RRs",
count, class, type);
ret = *lenp - buflen;
done:
if (found != NULL)
memput(found, (stalecount + 1) * sizeof *found);
if (tmpfound != NULL)
memput(tmpfound, (stalecount + 1) * sizeof *tmpfound);
*countp = count;
return (ret);
}
/*
* Do we want this data record based on the class and type?
*/
static int
wanted(const struct databuf *dp, int class, int type) {
const u_char *cp;
int coveredType;
time_t expiration;
#ifdef DEBUG
char pclass[15], ptype[15];
#endif
#ifdef DEBUG
strcpy(pclass, p_class(class));
strcpy(ptype, p_type(type));
ns_debug(ns_log_default, 3, "wanted(%#x, %s %s) [%s %s]",
dp, pclass, ptype,
p_class(dp->d_class), p_type(dp->d_type));
#endif
if (dp->d_class != class && class != C_ANY)
return (0);
/*
* Must check SIG for expiration below, other matches
* return OK here.
*/
if (type == dp->d_type && (type != T_SIG))
return (1);
/* For a T_ANY query, we do not want to return -ve $ed RRs. */
if (type == T_ANY && dp->d_rcode == NOERROR_NODATA)
return (0);
/* First, look at the type of RR. */
switch (dp->d_type) {
/* Cases to deal with:
T_ANY search, return all unexpired SIGs.
T_SIG search, return all unexpired SIGs.
T_<foo> search, return all unexp SIG <FOO>s.
*/
case T_SIG:
cp = dp->d_data;
GETSHORT(coveredType, cp);
cp += INT16SZ + INT32SZ; /* skip alg, labels, & orig TTL */
GETLONG(expiration,cp);
if (type == T_ANY || type == T_SIG) {
if (expiration > time(0))
return (1); /* Unexpired matching SIG */
}
return (0); /* We don't return this SIG. */
case T_ANY:
return (1);
case T_CNAME:
if (dp->d_rcode != NOERROR_NODATA)
return (1);
else
break;
}
/* OK, now look at the type of query. */
if (type == ns_t_any)
return (1);
else if (type == ns_t_mailb)
switch (dp->d_type) {
case T_MR:
case T_MB:
case T_MG:
case T_MINFO:
return (1);
}
else if (ns_t_xfr_p(type)) {
/*
* This is used to validate transfer requests, not
* generate transfer responses. Is there an SOA?
*/
if (dp->d_type == ns_t_soa && dp->d_zone != DB_Z_CACHE
&& (zones[dp->d_zone].z_flags & Z_AUTH))
return (1);
}
return (0);
}
static int
wantedsig(const struct databuf *dp, int class, int type) {
const u_char *cp;
int coveredType;
time_t expiration;
#ifdef DEBUG
char pclass[15], ptype[15];
#endif
#ifdef DEBUG
strcpy(pclass, p_class(class));
strcpy(ptype, p_type(type));
ns_debug(ns_log_default, 3, "wantedtsig(%#x, %s %s) [%s %s]",
dp, pclass, ptype,
p_class(dp->d_class), p_type(dp->d_type));
#endif
if (dp->d_class != class && class != C_ANY)
return (0);
if (dp->d_type != T_SIG || dp->d_rcode != 0)
return (0);
cp = dp->d_data;
GETSHORT(coveredType, cp);
cp += INT16SZ + INT32SZ; /* skip alg, labels, & orig TTL */
GETLONG(expiration,cp);
if (expiration < time(0))
return (0);
if (type == T_ANY || type == T_SIG || type == coveredType)
return (1);
if (type == ns_t_mailb) {
switch (coveredType) {
case T_MR:
case T_MB:
case T_MG:
case T_MINFO:
return (1);
}
}
return (0);
}
/*
* Add RR entries from dpp array to a query/response.
* Return the number of bytes added or negative the amount
* added if truncation occured. Typically you are
* adding NS records to a response.
*/
int
add_data(struct namebuf *np, struct databuf **dpp,
u_char *cp, int buflen, int *countp)
{
struct databuf *dp;
char dname[MAXDNAME];
int n, bytes;
bytes = *countp = 0;
getname(np, dname, sizeof(dname));
for (dp = *dpp++; dp != NULL; dp = *dpp++) {
if (stale(dp))
continue; /* ignore old cache entry */
if (dp->d_rcode)
continue;
if ((n = make_rr(dname, dp, cp, buflen, 1,
dnptrs, dnptrs_end, 0)) < 0)
return (-bytes); /* Truncation */
cp += n;
buflen -= n;
bytes += n;
(*countp)++;
}
return (bytes);
}
static void
rrsetadd(struct flush_set *flushset, const char *name, struct databuf *dp) {
struct flush_set *fs = flushset;
struct db_list *dbl;
while (fs->fs_name && (
ns_samename(fs->fs_name,name) != 1 ||
(fs->fs_class != dp->d_class) ||
(fs->fs_type != dp->d_type) ||
(fs->fs_cred != dp->d_cred))) {
fs++;
}
if (!fs->fs_name) {
fs->fs_name = savestr(name, 1);
fs->fs_class = dp->d_class;
fs->fs_type = dp->d_type;
fs->fs_cred = dp->d_cred;
fs->fs_list = NULL;
fs->fs_last = NULL;
}
dbl = (struct db_list *)memget(sizeof(struct db_list));
if (!dbl)
panic("rrsetadd: out of memory", NULL);
dbl->db_next = NULL;
dbl->db_dp = dp;
if (fs->fs_last == NULL)
fs->fs_list = dbl;
else
fs->fs_last->db_next = dbl;
fs->fs_last = dbl;
}
static int
ttlcheck(const char *name, struct db_list *dbl, int update) {
int type = dbl->db_dp->d_type;
int class = dbl->db_dp->d_class;
struct hashbuf *htp = hashtab;
const char *fname;
struct namebuf *np;
struct db_list *dbp = dbl;
struct databuf *dp;
u_int32_t ttl = 0; /* Make gcc happy. */
int first;
np = nlookup(name, &htp, &fname, 0);
if (np == NULL || fname != name || ns_wildcard(NAME(*np)))
return (1);
/* check that all the ttl's we have are the same, if not return 1 */
first = 1;
for (dp = np->n_data; dp != NULL; dp = dp->d_next) {
if (!match(dp, class, type))
continue;
if (first) {
/* we can't update zone data so return early */
if (dp->d_zone != DB_Z_CACHE)
return (0);
ttl = dp->d_ttl;
first = 0;
} else if (ttl != dp->d_ttl)
return (1);
}
/* there are no records of this type in the cache */
if (first)
return(1);
/*
* the ttls of all records we have in the cache are the same
* if the ttls differ in the new set we don't want it.
*/
/* check that all the ttl's we have are the same, if not return 0 */
first = 1;
while (dbp) {
if (first) {
ttl = dbp->db_dp->d_ttl;
first = 0;
} else if (ttl != dbp->db_dp->d_ttl) {
return(0);
}
dbp = dbp->db_next;
}
/* update ttl if required */
if (update) {
for (dp = np->n_data; dp != NULL; dp = dp->d_next) {
if (!match(dp, class, type))
continue;
if (dp->d_ttl > ttl)
break;
dp->d_ttl = ttl;
fixttl(dp);
}
}
return(1);
}
/*
* lookup rrset in table and compare to dbl
* tri state result
* -1: lookup failed
* 0: rrsets same
* 1: rrsets differ
*/
static int
rrsetcmp(char * name, struct db_list * dbl, struct hashbuf * table) {
int type = dbl->db_dp->d_type;
int class = dbl->db_dp->d_class;
struct hashbuf *htp = table;
const char *fname;
struct namebuf *np;
struct db_list *dbp = dbl;
struct databuf *dp;
int exists = 0;
np = nlookup(name, &htp, &fname, 0);
if (np == NULL || fname != name || ns_wildcard(NAME(*np))) {
ns_debug(ns_log_default, 3, "rrsetcmp: name not in database");
return (-1);
}
/* check that all entries in dbl are in the cache */
while (dbp) {
for (dp = np->n_data; dp != NULL; dp = dp->d_next) {
if (!match(dp, class, type))
continue;
exists = 1;
if (!db_cmp(dp, dbp->db_dp)
#ifdef NOADDITIONAL
&& ((dp->d_cred == dbp->db_dp->d_cred) ||
(dp->d_cred != DB_C_ADDITIONAL))
#endif
)
break;
}
if (!dp) {
ns_debug(ns_log_default, 3,
"rrsetcmp: %srecord%s in database",
exists ? "" : "no ", exists ? " not" : "s");
return (exists ? 1 : -1);
}
dbp = dbp->db_next;
}
/* Check that all cache entries are in the list. */
for (dp = np->n_data; dp != NULL; dp = dp->d_next) {
if (!match(dp, class, type))
continue;
#ifdef NCACHE
if (dp->d_rcode)
return (1);
#endif
dbp = dbl;
while (dbp) {
if (!db_cmp(dp, dbp->db_dp))
break;
dbp = dbp->db_next;
}
if (!dbp) {
ns_debug(ns_log_default, 3,
"rrsetcmp: record not in rrset");
return (1);
}
}
ns_debug(ns_log_default, 3, "rrsetcmp: rrsets matched");
return (0);
}
/*
* verify incoming answer against what we already have in the hints
* issue warnings / errors if differences detected.
*/
static void
check_hints(struct flush_set * flushset) {
struct zoneinfo *zp;
struct flush_set *fs;
struct db_list *dbp;
/* We don't use hints when in forward only mode */
if (NS_OPTION_P(OPTION_FORWARD_ONLY))
return;
/* find "." NS rrset and hence class */
for (fs = flushset; fs->fs_name != NULL; fs++) {
if ((fs->fs_name[0] != '\0') || (fs->fs_type != ns_t_ns))
continue;
/* see if we are a root server */
zp = find_zone(fs->fs_name, fs->fs_class);
if (zp != NULL &&
(zp->z_type == z_master || zp->z_type == z_slave))
return;
switch (rrsetcmp(fs->fs_name, fs->fs_list, fcachetab)) {
case -1:
ns_error(ns_log_default,
"check_hints: no NS records for class %d in hints",
fs->fs_class);
break;
case 1:
ns_warning(ns_log_default,
"check_hints: root NS list in hints for class %d does not match root NS list",
fs->fs_class);
break;
case 0:
break;
default:
ns_error(ns_log_default,
"check_hints: unexpected response from rrsetcmp");
break;
}
break;
}
if (fs->fs_name == NULL) /* no root NS records */
return;
dbp = fs->fs_list;
while (dbp) {
/* for each NS find A rrset in answer and check */
for (fs = flushset; fs->fs_name != NULL; fs++) {
if (ns_samename(fs->fs_name, (char *)dbp->db_dp->d_data) != 1
|| fs->fs_type != ns_t_a)
continue;
switch (rrsetcmp(fs->fs_name, fs->fs_list, fcachetab)) {
case -1:
ns_error(ns_log_default,
"check_hints: no A records for %s class %d in hints",
fs->fs_name[0] ? fs->fs_name : ".",
fs->fs_class);
break;
case 1:
ns_warning(ns_log_default,
"check_hints: A records for %s class %d do not match hint records",
fs->fs_name[0] ? fs->fs_name : ".",
fs->fs_class);
break;
case 0:
break;
default:
ns_error(ns_log_default,
"check_hints: unexpected response from rrsetcmp");
break;
}
break;
}
if (fs->fs_name == NULL)
ns_debug(ns_log_default, 2,
"check_hints: no A records for %s",
dbp->db_dp->d_data);
dbp = dbp->db_next;
}
}
static void
rrsetupdate(struct flush_set * flushset, int flags, struct sockaddr_in from,
int updatettl) {
struct flush_set *fs = flushset;
struct db_list *dbp, *odbp;
int n;
void *state = NULL;
while (fs->fs_name) {
ns_debug(ns_log_default, 2, "rrsetupdate: %s",
fs->fs_name[0] ? fs->fs_name : ".");
if ((n = rrsetcmp(fs->fs_name, fs->fs_list, hashtab)) &&
ttlcheck(fs->fs_name, fs->fs_list, 0)) {
if (n > 0)
flushrrset(fs, from);
dbp = fs->fs_list;
while (dbp) {
n = db_set_update(fs->fs_name, dbp->db_dp,
&state, flags,
&hashtab, from, NULL,
0, NULL);
ns_debug(ns_log_default, 3,
"rrsetupdate: %s %d",
fs->fs_name[0] ? fs->fs_name : ".",
n);
odbp = dbp;
dbp = dbp->db_next;
memput(odbp, sizeof *odbp);
}
ns_debug(ns_log_default, 3,
"rrsetupdate: %s %d",
fs->fs_name[0] ? fs->fs_name : ".", n);
} else {
if ((n == 0) && updatettl)
(void)ttlcheck(fs->fs_name,fs->fs_list, 1);
dbp = fs->fs_list;
while (dbp) {
db_freedata(dbp->db_dp);
odbp = dbp;
dbp = dbp->db_next;
memput(odbp, sizeof *odbp);
}
}
fs->fs_list = NULL;
fs++;
}
n = db_set_update(NULL, NULL, &state, flags, &hashtab, from,
NULL, 0, NULL);
}
static void
flushrrset(struct flush_set * fs, struct sockaddr_in from) {
struct databuf *dp;
int n;
ns_debug(ns_log_default, 2, "flushrrset(%s, %s, %s, %d)",
fs->fs_name[0]?fs->fs_name:".", p_type(fs->fs_type),
p_class(fs->fs_class), fs->fs_cred);
dp = savedata(fs->fs_class, fs->fs_type, 0, NULL, 0);
dp->d_zone = DB_Z_CACHE;
dp->d_cred = fs->fs_cred;
dp->d_clev = 0;
do {
n = db_update(fs->fs_name, dp, NULL, NULL, DB_DELETE, hashtab,
from);
ns_debug(ns_log_default, 3, "flushrrset: %d", n);
} while (n == OK);
db_freedata(dp);
}
static void
free_flushset(struct flush_set *flushset, int flushset_size) {
struct flush_set *fs;
for (fs = flushset; fs->fs_name != NULL; fs++)
freestr(fs->fs_name);
memput(flushset, flushset_size);
}
/*
* This is best thought of as a "cache invalidate" function.
* It is called whenever a piece of data is determined to have
* become invalid either through a timeout or a validation
* failure. It is better to have no information, than to
* have partial information you pass off as complete.
*/
void
delete_all(struct namebuf *np, int class, int type) {
struct databuf *dp, *pdp;
ns_debug(ns_log_default, 3, "delete_all(%#x:\"%s\" %s %s)",
np, NAME(*np), p_class(class), p_type(type));
pdp = NULL;
dp = np->n_data;
while (dp != NULL) {
if (dp->d_zone == DB_Z_CACHE && (dp->d_flags & DB_F_HINT) == 0
&& match(dp, class, type)) {
dp = rm_datum(dp, np, pdp, NULL);
continue;
}
pdp = dp;
dp = dp->d_next;
}
}
/* delete_stale(np)
* for all RRs associated with this name, check for staleness (& delete)
* arguments:
* np = pointer to namebuf to be cleaned.
* returns:
* number of RRs associated with this name.
* side effects:
* delete_all() can be called, freeing memory and relinking chains.
*/
int
delete_stale(np)
struct namebuf *np;
{
struct databuf *dp;
int count;
again:
count = 0;
for (dp = np->n_data; dp != NULL; dp = dp->d_next) {
if (dp->d_zone == DB_Z_CACHE && stale(dp)) {
delete_all(np, dp->d_class, dp->d_type);
goto again;
}
count++;
}
return (count);
}
/*
* Adjust answer message so that it fits in outlen. Set tc if required.
*
* If outlen = msglen, can be used to verify qdcount, ancount, nscount
* and arcount.
*
* return new length
*/
int
trunc_adjust(u_char *msg, int msglen, int outlen) {
register HEADER *hp;
u_int qdcount, ancount, nscount, arcount, dlen;
u_char *cp = msg, *cp1, *eom_in, *eom_out;
int n;
eom_in = msg + msglen;
eom_out = msg + outlen;
hp = (HEADER *)msg;
qdcount = ntohs(hp->qdcount);
ancount = ntohs(hp->ancount);
nscount = ntohs(hp->nscount);
arcount = ntohs(hp->arcount);
cp += HFIXEDSZ;
while ((qdcount || ancount || nscount || arcount) &&
cp < eom_in && cp < eom_out) {
cp1 = cp; /* use temporary in case we break */
n = dn_skipname(cp1, eom_in);
if (n < 0)
break;
cp1 += n + 2 * INT16SZ; /* type, class */
if (!qdcount) {
cp1 += INT32SZ; /* ttl */
if (cp1 + INT16SZ > eom_in)
break;
GETSHORT(dlen, cp1);
cp1 += dlen;
}
if (cp1 > eom_in || cp1 > eom_out)
break;
cp = cp1;
if (qdcount)
qdcount--;
else if (ancount)
ancount--;
else if (nscount)
nscount--;
else
arcount--;
}
if (qdcount || ancount || nscount || arcount) {
ns_debug(ns_log_default, 1,
"trunc_adjust:%s %d %d %d %d %d, %d %d %d %d %d",
hp->tc?" tc":"", msglen,
ntohs(hp->qdcount), ntohs(hp->ancount),
ntohs(hp->nscount), ntohs(hp->arcount),
cp-msg, qdcount, ancount, nscount, arcount);
hp->tc = 1;
hp->qdcount = htons(ntohs(hp->qdcount) - qdcount);
hp->ancount = htons(ntohs(hp->ancount) - ancount);
hp->nscount = htons(ntohs(hp->nscount) - nscount);
hp->arcount = htons(ntohs(hp->arcount) - arcount);
}
ENSURE(cp <= eom_out);
return (cp - msg);
}
/*
* mark the server "from" bad in the qp structure so it won't be retried.
*/
static void
mark_bad(struct qinfo *qp, struct sockaddr_in from) {
int i;
for (i = 0; i < (int)qp->q_naddr; i++)
if (ina_equal(qp->q_addr[i].ns_addr.sin_addr, from.sin_addr))
qp->q_addr[i].nretry = MAXRETRY;
}
static void
mark_lame(struct qinfo *qp, struct sockaddr_in from) {
int i;
for (i = 0; i < (int)qp->q_naddr; i++)
if (ina_equal(qp->q_addr[i].ns_addr.sin_addr, from.sin_addr) &&
qp->q_addr[i].ns != NULL) {
qp->q_addr[i].ns->d_flags |= DB_F_LAME;
db_lame_add(qp->q_domain,
(char*)qp->q_addr[i].ns->d_data,
tt.tv_sec + server_options->lame_ttl);
}
}
/*
* Retry the message if and only if from matches where the query was
* last sent to. The code does not handle responses sent from the
* wrong interface an a multihomed server.
*/
static void
fast_retry(struct qinfo *qp, struct sockaddr_in from) {
if (ina_equal(qp->q_addr[qp->q_curaddr].ns_addr.sin_addr,
from.sin_addr))
retry(qp);
}
static void
add_related_additional(char *name) {
int i;
if (num_related >= MAX_RELATED - 1)
return;
for (i = 0; i < num_related; i++)
if (ns_samename(name, related[i]) == 1) {
freestr(name);
return;
}
related[num_related++] = name;
}
static void
free_related_additional() {
int i;
for (i = 0; i < num_related; i++)
freestr(related[i]);
num_related = 0;
}
static int
related_additional(char *name) {
int i;
for (i = 0; i < num_related; i++)
if (ns_samename(name, related[i]) == 1)
return (1);
return (0);
}
static void
freestr_maybe(char **tname) {
if (tname == NULL || *tname == NULL)
return;
freestr(*tname);
*tname = NULL;
}
/*
* Match a request namebuf against the configured rrset-order info. First
* match wins. There is an implicit '*.' at the front to the ordering names.
*/
static enum ordering
match_order(const struct namebuf *np, int class, int type) {
rrset_order_list orders = server_options->ordering;
rrset_order_element roe;
if (orders == NULL)
return (DEFAULT_ORDERING);
for (roe = orders->first ; roe != NULL ; roe = roe->next) {
if (roe->class != C_ANY && roe->class != class)
continue;
if (roe->type != T_ANY && roe->type != type)
continue;
if (match_name(np, roe->name, strlen(roe->name)) == 0) {
return (roe->order);
}
}
/* none matched so use default */
return (DEFAULT_ORDERING);
}
/* Do a simple compare of the NP data against the given NAME, recursively
* looking at the NP parent if necessary. NAMELEN is the length of the NAME
* that needs to be matched. Matching happen from right to left. Returns -1
* on failure, on success the index of the first character of the matched
* portion of the string is returned. In the first level call a return
* value of 0 is of interest.
*/
static int
match_name(const struct namebuf *np, const char *name, size_t namelen)
{
int matched ;
if (name[0] == '*' && name[1] == '\0')
return 0;
if (np->n_parent != NULL) { /* recurse to end of np list */
matched = match_name(np->n_parent,name,namelen);
} else {
matched = namelen;
}
if (matched > 0) {
int labellen = NAMELEN(*np);
char pch;
const char *start;
if (labellen > matched) {
return -1;
} else if (labellen < matched) {
/* string is longer than this namebuf's data, so
make sure there's a period before the end of the
match so we don't just match a suffix. */
start = name + (matched - labellen);
pch = start[-1];
if (pch != '.') {
return -1;
}
} else {
start = name ;
}
if (strncasecmp(start, NAME(*np), labellen) == 0) {
/* looking good. tell our caller what portion of
the tail of string has been matched */
if (start == name)
return (0) ;
else
return (start - name - 1); /* matched '.' too */
} else {
return (-1);
}
}
return (matched);
}