3178 lines
75 KiB
C
3178 lines
75 KiB
C
/* $NetBSD: ns_config.c,v 1.6 2001/05/17 22:59:39 itojun Exp $ */
|
|
|
|
#if !defined(lint) && !defined(SABER)
|
|
static const char rcsid[] = "Id: ns_config.c,v 8.121 2001/02/08 02:05:53 marka Exp";
|
|
#endif /* not lint */
|
|
|
|
/*
|
|
* 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.
|
|
*/
|
|
|
|
/*
|
|
* Portions Copyright (c) 1999 by Check Point Software Technologies, Inc.
|
|
*
|
|
* 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 Check Point Software Technologies Incorporated 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 CHECK POINT SOFTWARE TECHNOLOGIES
|
|
* INCORPORATED DISCLAIMS ALL WARRANTIES WITH REGARD TO THIS SOFTWARE,
|
|
* INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS.
|
|
* IN NO EVENT SHALL CHECK POINT SOFTWARE TECHNOLOGIES INCORPRATED
|
|
* 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/stat.h>
|
|
#include <sys/un.h>
|
|
|
|
#include <netinet/in.h>
|
|
#include <arpa/nameser.h>
|
|
#include <arpa/inet.h>
|
|
|
|
#include <ctype.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 <unistd.h>
|
|
#include <fcntl.h>
|
|
|
|
#include <isc/eventlib.h>
|
|
#include <isc/logging.h>
|
|
#include <isc/memcluster.h>
|
|
|
|
#include <isc/dst.h>
|
|
|
|
#include "port_after.h"
|
|
|
|
#ifdef HAVE_GETRUSAGE /* XXX */
|
|
#include <sys/resource.h>
|
|
#endif
|
|
|
|
#include "named.h"
|
|
#include "ns_parseutil.h"
|
|
|
|
/* Private. */
|
|
|
|
static int tmpnum = 0;
|
|
static int config_initialized = 0;
|
|
|
|
static int need_logging_free = 0;
|
|
static int default_logging_installed;
|
|
|
|
static int options_installed = 0;
|
|
static int logging_installed = 0;
|
|
static int default_options_installed;
|
|
static int initial_configuration = 1;
|
|
|
|
static char **logging_categories;
|
|
static char *current_pid_filename = NULL;
|
|
|
|
#define ZONE_SYM_TABLE_SIZE 4973
|
|
static symbol_table zone_symbol_table;
|
|
|
|
/* Zones */
|
|
|
|
void
|
|
free_zone_timerinfo(struct zoneinfo *zp) {
|
|
if (zp->z_timerinfo != NULL) {
|
|
freestr(zp->z_timerinfo->name);
|
|
memput(zp->z_timerinfo, sizeof *zp->z_timerinfo);
|
|
zp->z_timerinfo = NULL;
|
|
} else
|
|
ns_error(ns_log_config, "timer for zone '%s' had no timerinfo",
|
|
zp->z_origin);
|
|
}
|
|
|
|
void
|
|
free_zone_contents(struct zoneinfo *zp, int undefine_sym) {
|
|
INSIST(zp != NULL);
|
|
|
|
if (undefine_sym)
|
|
undefine_symbol(zone_symbol_table, zp->z_origin, zp->z_class);
|
|
if (zp->z_flags & Z_TIMER_SET) {
|
|
free_zone_timerinfo(zp);
|
|
if (evClearTimer(ev, zp->z_timer) < 0)
|
|
ns_error(ns_log_config,
|
|
"evClearTimer for zone '%s' failed in ns_init: %s",
|
|
zp->z_origin,
|
|
strerror(errno));
|
|
}
|
|
if (zp->z_origin != NULL)
|
|
freestr(zp->z_origin);
|
|
zp->z_origin = NULL;
|
|
if (zp->z_source != NULL)
|
|
freestr(zp->z_source);
|
|
zp->z_source = NULL;
|
|
if (zp->z_ixfr_base != NULL)
|
|
freestr(zp->z_ixfr_base);
|
|
zp->z_ixfr_base = NULL;
|
|
if (zp->z_ixfr_tmp != NULL)
|
|
freestr(zp->z_ixfr_tmp);
|
|
zp->z_ixfr_tmp = NULL;
|
|
if (zp->z_update_acl != NULL)
|
|
free_ip_match_list(zp->z_update_acl);
|
|
zp->z_update_acl = NULL;
|
|
if (zp->z_query_acl != NULL)
|
|
free_ip_match_list(zp->z_query_acl);
|
|
zp->z_query_acl = NULL;
|
|
if (zp->z_transfer_acl != NULL)
|
|
free_ip_match_list(zp->z_transfer_acl);
|
|
zp->z_transfer_acl = NULL;
|
|
#ifdef BIND_UPDATE
|
|
if (zp->z_updatelog != NULL)
|
|
freestr(zp->z_updatelog);
|
|
zp->z_updatelog = NULL;
|
|
#endif /* BIND_UPDATE */
|
|
#ifdef BIND_NOTIFY
|
|
if (zp->z_also_notify != NULL)
|
|
memput(zp->z_also_notify,
|
|
zp->z_notify_count * sizeof *zp->z_also_notify);
|
|
zp->z_also_notify = NULL;
|
|
#endif
|
|
block_signals();
|
|
if (LINKED(zp, z_reloadlink))
|
|
UNLINK(reloadingzones, zp, z_reloadlink);
|
|
unblock_signals();
|
|
}
|
|
|
|
static void
|
|
release_zone(struct zoneinfo *zp) {
|
|
INSIST(zp != NULL);
|
|
|
|
free_zone_contents(zp, 0);
|
|
memput(zp, sizeof *zp);
|
|
}
|
|
|
|
struct zoneinfo *
|
|
find_zone(const char *name, int class) {
|
|
struct zoneinfo *zp;
|
|
symbol_value value;
|
|
|
|
ns_debug(ns_log_config, 3, "find_zone(%s, %d)",
|
|
*name ? name : ".", class);
|
|
if (lookup_symbol(zone_symbol_table, name, class, &value)) {
|
|
INSIST(value.integer >= 0 && value.integer < nzones);
|
|
ns_debug(ns_log_config, 3, "find_zone: existing zone %d",
|
|
value.integer);
|
|
zp = &zones[value.integer];
|
|
return (zp);
|
|
}
|
|
ns_debug(ns_log_config, 3, "find_zone: unknown zone");
|
|
return (NULL);
|
|
}
|
|
|
|
static struct zoneinfo *
|
|
new_zone(int class, int type) {
|
|
struct zoneinfo *zp;
|
|
|
|
if (EMPTY(freezones))
|
|
make_new_zones();
|
|
|
|
zp = HEAD(freezones);
|
|
UNLINK(freezones, zp, z_freelink);
|
|
return (zp);
|
|
}
|
|
|
|
/*
|
|
* Check out a zoneinfo structure and return non-zero if it's OK.
|
|
*/
|
|
static int
|
|
validate_zone(struct zoneinfo *zp) {
|
|
char filename[MAXPATHLEN+1];
|
|
|
|
/* Check name */
|
|
if (!res_dnok(zp->z_origin)) {
|
|
ns_error(ns_log_config, "invalid zone name '%s'",
|
|
zp->z_origin);
|
|
return (0);
|
|
}
|
|
|
|
/* Check class */
|
|
if (zp->z_class == C_ANY || zp->z_class == C_NONE) {
|
|
ns_error(ns_log_config, "invalid class %d for zone '%s'",
|
|
zp->z_class, zp->z_origin);
|
|
return (0);
|
|
}
|
|
|
|
/* Check type. */
|
|
if (zp->z_type == 0) {
|
|
ns_error(ns_log_config, "no type specified for zone '%s'",
|
|
zp->z_origin);
|
|
return (0);
|
|
}
|
|
if (zp->z_type == z_cache && ns_samename(zp->z_origin, "") != 1) {
|
|
ns_error(ns_log_config,
|
|
"only the root zone may be a cache zone (zone '%s')",
|
|
zp->z_origin);
|
|
return (0);
|
|
}
|
|
if (zp->z_type == z_hint && ns_samename(zp->z_origin, "") != 1) {
|
|
ns_error(ns_log_config,
|
|
"only the root zone may be a hint zone (zone '%s')",
|
|
zp->z_origin);
|
|
return (0);
|
|
}
|
|
|
|
/* Check filename. */
|
|
if (zp->z_type == z_master && zp->z_source == NULL) {
|
|
ns_error(ns_log_config,
|
|
"'file' statement missing for master zone %s",
|
|
zp->z_origin);
|
|
return (0);
|
|
}
|
|
/*
|
|
* XXX We should run filename through an OS-specific
|
|
* validator here.
|
|
*/
|
|
if (zp->z_source != NULL &&
|
|
strlen(zp->z_source) > MAXPATHLEN) {
|
|
ns_error(ns_log_config, "filename too long for zone '%s'",
|
|
zp->z_origin);
|
|
return (0);
|
|
}
|
|
|
|
if (zp->z_ixfr_base != NULL && strlen(zp->z_ixfr_base) > MAXPATHLEN) {
|
|
ns_error(ns_log_config, "ixfr filename too long for zone '%s'",
|
|
zp->z_origin);
|
|
return (0);
|
|
}
|
|
if (zp->z_ixfr_tmp != NULL && strlen(zp->z_ixfr_tmp) > MAXPATHLEN) {
|
|
ns_error(ns_log_config, "tmp ixfr filename too long for zone '%s'",
|
|
zp->z_origin);
|
|
return (0);
|
|
}
|
|
|
|
/* Check masters */
|
|
if (zp->z_addrcnt != 0) {
|
|
if (zp->z_type == z_master || zp->z_type == z_hint ||
|
|
zp->z_type == z_cache) {
|
|
ns_error(ns_log_config,
|
|
"'masters' statement present for %s zone '%s'",
|
|
(zp->z_type == z_master) ? "master" :
|
|
(zp->z_type == z_hint) ? "hint" : "cache",
|
|
zp->z_origin);
|
|
return (0);
|
|
}
|
|
} else {
|
|
if (zp->z_type == z_slave || zp->z_type == z_stub) {
|
|
ns_error(ns_log_config,
|
|
"no 'masters' statement for non-master zone '%s'",
|
|
zp->z_origin);
|
|
return (0);
|
|
}
|
|
}
|
|
|
|
/* Check allow-update and allow-transfer. */
|
|
if (zp->z_update_acl || zp->z_transfer_acl) {
|
|
if (zp->z_type != z_master && zp->z_type != z_slave) {
|
|
ns_error(ns_log_config,
|
|
"'allow-{update,transfer}' option for non-{master,slave} zone '%s'",
|
|
zp->z_origin);
|
|
return (0);
|
|
}
|
|
}
|
|
|
|
/* Check allow-query. */
|
|
if (zp->z_query_acl) {
|
|
if (zp->z_type != z_master &&
|
|
zp->z_type != z_slave &&
|
|
zp->z_type != z_stub) {
|
|
ns_error(ns_log_config,
|
|
"'allow-query' option for non-{master,slave,stub} zone '%s'",
|
|
zp->z_origin);
|
|
return (0);
|
|
}
|
|
}
|
|
|
|
#ifdef BIND_NOTIFY
|
|
/* Check notify */
|
|
if (zp->z_notify != znotify_use_default) {
|
|
if (zp->z_type != z_master && zp->z_type != z_slave) {
|
|
ns_error(ns_log_config,
|
|
"'notify' given for non-master, non-slave zone '%s'",
|
|
zp->z_origin);
|
|
return (0);
|
|
}
|
|
}
|
|
|
|
/* Check also-notify */
|
|
if (zp->z_notify_count != 0) {
|
|
if (zp->z_type != z_master && zp->z_type != z_slave) {
|
|
ns_error(ns_log_config,
|
|
"'also-notify' given for non-master, non-slave zone '%s'",
|
|
zp->z_origin);
|
|
return (0);
|
|
}
|
|
}
|
|
#endif
|
|
|
|
#ifdef BIND_UPDATE
|
|
/* XXX need more checking here */
|
|
if (!zp->z_updatelog && zp->z_source) {
|
|
/* XXX OS-specific filename validation here */
|
|
if ((strlen(zp->z_source) + (sizeof ".log" - 1)) >
|
|
MAXPATHLEN) {
|
|
ns_error(ns_log_config,
|
|
"filename too long for dynamic zone '%s'",
|
|
zp->z_origin);
|
|
return (0);
|
|
}
|
|
/* this sprintf() is now safe */
|
|
sprintf(filename, "%s.log", zp->z_source);
|
|
zp->z_updatelog = savestr(filename, 1);
|
|
}
|
|
|
|
/* Check forward */
|
|
if (zp->z_optset & OPTION_FORWARD_ONLY) {
|
|
if (zp->z_type == z_hint) {
|
|
ns_error(ns_log_config,
|
|
"'forward' given for hint zone '%s'",
|
|
zp->z_origin);
|
|
return (0);
|
|
}
|
|
}
|
|
/* Check forwarders */
|
|
if (zp->z_fwdtab) {
|
|
if (zp->z_type == z_hint) {
|
|
ns_error(ns_log_config,
|
|
"'forwarders' given for hint zone '%s'",
|
|
zp->z_origin);
|
|
return (0);
|
|
}
|
|
}
|
|
|
|
if (zp->z_type == z_master) {
|
|
if (!zp->z_soaincrintvl)
|
|
zp->z_soaincrintvl = SOAINCRINTVL;
|
|
if (!zp->z_dumpintvl)
|
|
zp->z_dumpintvl = DUMPINTVL;
|
|
if (!zp->z_deferupdcnt)
|
|
zp->z_deferupdcnt = DEFERUPDCNT;
|
|
}
|
|
#endif /* BIND_UPDATE */
|
|
|
|
if (!zp->z_ixfr_base && zp->z_source) {
|
|
/* XXX OS-specific filename validation here */
|
|
if ((strlen(zp->z_source) + (sizeof ".ixfr" - 1)) >
|
|
MAXPATHLEN) {
|
|
ns_error(ns_log_config,
|
|
"filename too long for dynamic zone '%s'",
|
|
zp->z_origin);
|
|
return (0);
|
|
}
|
|
/* this sprintf() is now safe */
|
|
sprintf(filename, "%s.ixfr", zp->z_source);
|
|
zp->z_ixfr_base = savestr(filename, 1);
|
|
}
|
|
if (!zp->z_ixfr_tmp && zp->z_source) {
|
|
/* XXX OS-specific filename validation here */
|
|
if ((strlen(zp->z_source) + (sizeof ".ixfr.tmp" - 1)) >
|
|
MAXPATHLEN) {
|
|
ns_error(ns_log_config,
|
|
"filename too long for dynamic zone '%s'",
|
|
zp->z_origin);
|
|
return (0);
|
|
}
|
|
/* this sprintf() is now safe */
|
|
sprintf(filename, "%s.ixfr.tmp", zp->z_source);
|
|
zp->z_ixfr_tmp = savestr(filename, 1);
|
|
}
|
|
|
|
return (1);
|
|
}
|
|
|
|
/*
|
|
* Start building a new zoneinfo structure. Returns an opaque
|
|
* zone_config suitable for use by the parser.
|
|
*/
|
|
zone_config
|
|
begin_zone(char *name, int class) {
|
|
zone_config zh;
|
|
struct zoneinfo *zp;
|
|
|
|
/*
|
|
* require: name is canonical, class is a valid class
|
|
*/
|
|
|
|
ns_debug(ns_log_config, 3, "begin_zone('%s', %d)",
|
|
(*name == '\0') ? "." : name, class);
|
|
|
|
zp = (struct zoneinfo *)memget(sizeof (struct zoneinfo));
|
|
if (zp == NULL)
|
|
panic("memget failed in begin_zone", NULL);
|
|
memset(zp, 0, sizeof (struct zoneinfo));
|
|
zp->z_origin = name;
|
|
zp->z_class = class;
|
|
zp->z_checknames = not_set;
|
|
if (server_options->flags & OPTION_MAINTAIN_IXFR_BASE)
|
|
zp->z_maintain_ixfr_base = 1;
|
|
else
|
|
zp->z_maintain_ixfr_base = 0;
|
|
zp->z_max_log_size_ixfr = server_options->max_log_size_ixfr;
|
|
zh.opaque = zp;
|
|
return (zh);
|
|
}
|
|
|
|
/*
|
|
* Merge new configuration information into an existing zone. The
|
|
* new zoneinfo must be valid.
|
|
*/
|
|
static void
|
|
update_zone_info(struct zoneinfo *zp, struct zoneinfo *new_zp) {
|
|
char buf[MAXPATHLEN+1];
|
|
int i;
|
|
|
|
INSIST(zp != NULL);
|
|
INSIST(new_zp != NULL);
|
|
|
|
ns_debug(ns_log_config, 1, "update_zone_info('%s', %d)",
|
|
(*new_zp->z_origin == '\0') ? "." : new_zp->z_origin,
|
|
new_zp->z_type);
|
|
|
|
#ifdef BIND_UPDATE
|
|
/*
|
|
* A dynamic master zone that's becoming non-dynamic may need to be
|
|
* dumped before we start the update.
|
|
*/
|
|
if ((zp->z_flags & Z_DYNAMIC) && !(new_zp->z_flags & Z_DYNAMIC) &&
|
|
((zp->z_flags & Z_NEED_SOAUPDATE) ||
|
|
(zp->z_flags & Z_NEED_DUMP)))
|
|
(void) zonedump(zp, ISNOTIXFR);
|
|
#endif
|
|
|
|
/*
|
|
* First do the simple stuff, making sure to free
|
|
* any data that was dynamically allocated.
|
|
*/
|
|
if (zp->z_origin != NULL)
|
|
freestr(zp->z_origin);
|
|
zp->z_origin = new_zp->z_origin;
|
|
new_zp->z_origin = NULL;
|
|
zp->z_maintain_ixfr_base = new_zp->z_maintain_ixfr_base;
|
|
zp->z_max_log_size_ixfr = new_zp->z_max_log_size_ixfr;
|
|
zp->z_class = new_zp->z_class;
|
|
zp->z_type = new_zp->z_type;
|
|
zp->z_checknames = new_zp->z_checknames;
|
|
for (i = 0; i < new_zp->z_addrcnt; i++)
|
|
zp->z_addr[i] = new_zp->z_addr[i];
|
|
zp->z_addrcnt = new_zp->z_addrcnt;
|
|
if (zp->z_update_acl)
|
|
free_ip_match_list(zp->z_update_acl);
|
|
zp->z_update_acl = new_zp->z_update_acl;
|
|
new_zp->z_update_acl = NULL;
|
|
if (zp->z_query_acl)
|
|
free_ip_match_list(zp->z_query_acl);
|
|
zp->z_query_acl = new_zp->z_query_acl;
|
|
new_zp->z_query_acl = NULL;
|
|
zp->z_axfr_src = new_zp->z_axfr_src;
|
|
if (zp->z_transfer_acl)
|
|
free_ip_match_list(zp->z_transfer_acl);
|
|
zp->z_transfer_acl = new_zp->z_transfer_acl;
|
|
new_zp->z_transfer_acl = NULL;
|
|
zp->z_max_transfer_time_in = new_zp->z_max_transfer_time_in;
|
|
#ifdef BIND_NOTIFY
|
|
zp->z_notify = new_zp->z_notify;
|
|
if (zp->z_also_notify)
|
|
memput(zp->z_also_notify,
|
|
zp->z_notify_count * sizeof *zp->z_also_notify);
|
|
zp->z_also_notify = new_zp->z_also_notify;
|
|
zp->z_notify_count = new_zp->z_notify_count;
|
|
new_zp->z_also_notify = NULL;
|
|
new_zp->z_notify_count = 0;
|
|
#endif
|
|
if ((new_zp->z_flags & Z_FORWARD_SET) != 0)
|
|
zp->z_flags |= Z_FORWARD_SET;
|
|
else
|
|
zp->z_flags &= ~Z_FORWARD_SET;
|
|
if (zp->z_fwdtab != NULL)
|
|
free_forwarders(zp->z_fwdtab);
|
|
zp->z_fwdtab = new_zp->z_fwdtab;
|
|
new_zp->z_fwdtab = NULL;
|
|
|
|
zp->z_dialup = new_zp->z_dialup;
|
|
zp->z_options = new_zp->z_options;
|
|
zp->z_optset = new_zp->z_optset;
|
|
|
|
#ifdef BIND_UPDATE
|
|
if (new_zp->z_flags & Z_DYNAMIC)
|
|
zp->z_flags |= Z_DYNAMIC;
|
|
else
|
|
zp->z_flags &= ~Z_DYNAMIC;
|
|
zp->z_soaincrintvl = new_zp->z_soaincrintvl;
|
|
zp->z_dumpintvl = new_zp->z_dumpintvl;
|
|
zp->z_deferupdcnt = new_zp->z_deferupdcnt;
|
|
if (zp->z_updatelog)
|
|
freestr(zp->z_updatelog);
|
|
zp->z_updatelog = new_zp->z_updatelog;
|
|
new_zp->z_updatelog = NULL;
|
|
#endif /* BIND_UPDATE */
|
|
zp->z_port = new_zp->z_port;
|
|
|
|
/*
|
|
* Now deal with files.
|
|
*/
|
|
switch (zp->z_type) {
|
|
case z_cache:
|
|
ns_panic(ns_log_config, 1, "impossible condition");
|
|
break;
|
|
case z_hint:
|
|
ns_debug(ns_log_config, 1, "source = %s", new_zp->z_source);
|
|
zp->z_refresh = 0; /* No dumping. */
|
|
if (zp->z_source != NULL &&
|
|
strcmp(new_zp->z_source, zp->z_source) == 0 &&
|
|
(reconfiging || !zonefile_changed_p(zp))) {
|
|
ns_debug(ns_log_config, 1, "cache is up to date");
|
|
break;
|
|
}
|
|
|
|
/* File has changed, or hasn't been loaded yet. */
|
|
if (zp->z_source) {
|
|
freestr(zp->z_source);
|
|
ns_stopxfrs(zp);
|
|
purge_zone(zp->z_origin, fcachetab, zp->z_class);
|
|
}
|
|
zp->z_source = new_zp->z_source;
|
|
new_zp->z_source = NULL;
|
|
|
|
if (zp->z_ixfr_base)
|
|
freestr(zp->z_ixfr_base);
|
|
zp->z_ixfr_base = new_zp->z_ixfr_base;
|
|
new_zp->z_ixfr_base = NULL;
|
|
|
|
if (zp->z_ixfr_tmp)
|
|
freestr(zp->z_ixfr_tmp);
|
|
zp->z_ixfr_tmp = new_zp->z_ixfr_tmp;
|
|
new_zp->z_ixfr_tmp = NULL;
|
|
|
|
ns_debug(ns_log_config, 1, "reloading hint zone");
|
|
(void) db_load(zp->z_source, zp->z_origin, zp, NULL,
|
|
ISNOTIXFR);
|
|
break;
|
|
|
|
case z_master:
|
|
ns_debug(ns_log_config, 1, "source = %s", new_zp->z_source);
|
|
/*
|
|
* If we've loaded this file, and the file hasn't changed
|
|
* then there's no need to reload.
|
|
*/
|
|
if (zp->z_source != NULL &&
|
|
strcmp(new_zp->z_source, zp->z_source) == 0 &&
|
|
(reconfiging || !zonefile_changed_p(zp))) {
|
|
ns_debug(ns_log_config, 1, "zone is up to date");
|
|
break;
|
|
}
|
|
#ifdef BIND_UPDATE
|
|
if (zp->z_source && (zp->z_flags & Z_DYNAMIC))
|
|
ns_warning(ns_log_config,
|
|
"source file of dynamic zone '%s' has changed",
|
|
zp->z_origin);
|
|
|
|
primary_reload:
|
|
#endif /* BIND_UPDATE */
|
|
if (zp->z_source != NULL)
|
|
freestr(zp->z_source);
|
|
zp->z_source = new_zp->z_source;
|
|
new_zp->z_source = NULL;
|
|
|
|
if (zp->z_ixfr_base != NULL)
|
|
freestr(zp->z_ixfr_base);
|
|
zp->z_ixfr_base = new_zp->z_ixfr_base;
|
|
new_zp->z_ixfr_base = NULL;
|
|
|
|
if (zp->z_ixfr_tmp != NULL)
|
|
freestr(zp->z_ixfr_tmp);
|
|
zp->z_ixfr_tmp = new_zp->z_ixfr_tmp;
|
|
new_zp->z_ixfr_tmp = NULL;
|
|
|
|
if (reload_master(zp) == 1) {
|
|
/*
|
|
* Note that going to primary_reload
|
|
* unconditionally reloads the zone.
|
|
*/
|
|
new_zp->z_source = savestr(zp->z_source, 1);
|
|
new_zp->z_ixfr_base = savestr(zp->z_ixfr_base, 1);
|
|
new_zp->z_ixfr_tmp = savestr(zp->z_ixfr_tmp, 1);
|
|
goto primary_reload;
|
|
}
|
|
break;
|
|
|
|
case z_slave:
|
|
#ifdef STUBS
|
|
case z_stub:
|
|
#endif
|
|
ns_debug(ns_log_config, 1, "addrcnt = %d", zp->z_addrcnt);
|
|
if (!new_zp->z_source) {
|
|
/*
|
|
* We will always transfer this zone again
|
|
* after a reload.
|
|
*/
|
|
sprintf(buf, "NsTmp%ld.%d", (long)getpid(), tmpnum++);
|
|
new_zp->z_source = savestr(buf, 1);
|
|
zp->z_flags |= Z_TMP_FILE;
|
|
} else
|
|
zp->z_flags &= ~Z_TMP_FILE;
|
|
/*
|
|
* If we had a backup file name, and it was changed,
|
|
* free old zone and start over. If we don't have
|
|
* current zone contents, try again now in case
|
|
* we have a new server on the list.
|
|
*/
|
|
if (zp->z_source != NULL &&
|
|
(strcmp(new_zp->z_source, zp->z_source) != 0 ||
|
|
((!reconfiging) && zonefile_changed_p(zp)))) {
|
|
ns_debug(ns_log_config, 1,
|
|
"backup file changed or missing");
|
|
freestr(zp->z_source);
|
|
zp->z_source = NULL;
|
|
zp->z_serial = 0; /* force xfer */
|
|
ns_stopxfrs(zp);
|
|
/*
|
|
* We only need to reload if we have ever
|
|
* successfully transferred the zone.
|
|
*/
|
|
if ((zp->z_flags & Z_AUTH) != 0) {
|
|
zp->z_flags &= ~Z_AUTH;
|
|
/*
|
|
* Purge old data and mark the parent for
|
|
* reloading so that NS records are present
|
|
* during the zone transfer.
|
|
*/
|
|
do_reload(zp->z_origin, zp->z_type,
|
|
zp->z_class, 1);
|
|
}
|
|
}
|
|
if (zp->z_source == NULL) {
|
|
zp->z_source = new_zp->z_source;
|
|
new_zp->z_source = NULL;
|
|
}
|
|
|
|
if (zp->z_ixfr_base != NULL)
|
|
freestr(zp->z_ixfr_base);
|
|
zp->z_ixfr_base = new_zp->z_ixfr_base;
|
|
new_zp->z_ixfr_base = NULL;
|
|
|
|
if (zp->z_ixfr_tmp != NULL)
|
|
freestr(zp->z_ixfr_tmp);
|
|
zp->z_ixfr_tmp = new_zp->z_ixfr_tmp;
|
|
new_zp->z_ixfr_tmp = NULL;
|
|
|
|
if ((!noexpired || ((zp->z_flags & Z_EXPIRED) == 0)) &&
|
|
((zp->z_flags & Z_AUTH) == 0))
|
|
zoneinit(zp);
|
|
else {
|
|
/*
|
|
** Force slave to try transfer soon after SIGHUP.
|
|
*/
|
|
if ((zp->z_flags & (Z_QSERIAL|Z_XFER_RUNNING)) == 0 &&
|
|
reloading && !reconfiging) {
|
|
qserial_retrytime(zp, tt.tv_sec);
|
|
sched_zone_maint(zp);
|
|
}
|
|
}
|
|
break;
|
|
case z_forward:
|
|
/*
|
|
* We don't know if the forwarder's list has changed
|
|
* so just purge the cache. In the future we may want
|
|
* see if the forwarders list has changed and only
|
|
* do this then.
|
|
*/
|
|
clean_cache_from(zp->z_origin, hashtab);
|
|
break;
|
|
}
|
|
if ((zp->z_flags & Z_FOUND) != 0 && /* already found? */
|
|
(zp - zones) != DB_Z_CACHE) /* cache never sets Z_FOUND */
|
|
ns_error(ns_log_config, "Zone \"%s\" declared more than once",
|
|
zp->z_origin);
|
|
zp->z_flags |= Z_FOUND;
|
|
ns_debug(ns_log_config, 1,
|
|
"zone[%d] type %d: '%s' z_time %lu, z_refresh %u",
|
|
zp-zones, zp->z_type,
|
|
*(zp->z_origin) == '\0' ? "." : zp->z_origin,
|
|
(u_long)zp->z_time, zp->z_refresh);
|
|
}
|
|
|
|
/*
|
|
* Finish constructing a new zone. If valid, the constructed zone is
|
|
* merged into the zone database. The zone_config used is invalid after
|
|
* end_zone() completes.
|
|
*/
|
|
void
|
|
end_zone(zone_config zh, int should_install) {
|
|
struct zoneinfo *zp, *new_zp;
|
|
char *zname;
|
|
symbol_value value;
|
|
|
|
new_zp = zh.opaque;
|
|
INSIST(new_zp != NULL);
|
|
|
|
zname = (new_zp->z_origin[0] == '\0') ? "." : new_zp->z_origin;
|
|
ns_debug(ns_log_config, 3, "end_zone('%s', %d)", zname,
|
|
should_install);
|
|
|
|
if (!should_install) {
|
|
release_zone(new_zp);
|
|
return;
|
|
}
|
|
if (!validate_zone(new_zp)) {
|
|
ns_error(ns_log_config,
|
|
"zone '%s' did not validate, skipping", zname);
|
|
release_zone(new_zp);
|
|
return;
|
|
}
|
|
zp = find_zone(new_zp->z_origin, new_zp->z_class);
|
|
if (zp != NULL && zp->z_type != new_zp->z_type) {
|
|
remove_zone(zp, "redefined");
|
|
zp = NULL;
|
|
}
|
|
if (zp == NULL) {
|
|
zp = new_zone(new_zp->z_class, new_zp->z_type);
|
|
INSIST(zp != NULL);
|
|
value.integer = (zp - zones);
|
|
define_symbol(zone_symbol_table, savestr(new_zp->z_origin, 1),
|
|
new_zp->z_class, value, SYMBOL_FREE_KEY);
|
|
}
|
|
ns_debug(ns_log_config, 5, "zone '%s', type = %d, class = %d", zname,
|
|
new_zp->z_type, new_zp->z_class);
|
|
if (new_zp->z_source != NULL)
|
|
ns_debug(ns_log_config, 5, " file = %s", new_zp->z_source);
|
|
ns_debug(ns_log_config, 5, " checknames = %d", new_zp->z_checknames);
|
|
if (new_zp->z_addrcnt != 0) {
|
|
int i;
|
|
|
|
ns_debug(ns_log_config, 5, " masters:");
|
|
for (i = 0; i < new_zp->z_addrcnt; i++)
|
|
ns_debug(ns_log_config, 5, " %s",
|
|
inet_ntoa(new_zp->z_addr[i]));
|
|
}
|
|
|
|
update_zone_info(zp, new_zp);
|
|
release_zone(new_zp);
|
|
zh.opaque = NULL;
|
|
}
|
|
|
|
int
|
|
set_zone_type(zone_config zh, int type) {
|
|
struct zoneinfo *zp;
|
|
|
|
zp = zh.opaque;
|
|
INSIST(zp != NULL);
|
|
|
|
/* Fail if type already set for this zone */
|
|
if (zp->z_type != 0)
|
|
return (0);
|
|
zp->z_type = type;
|
|
return (1);
|
|
}
|
|
|
|
int
|
|
set_zone_filename(zone_config zh, char *filename) {
|
|
struct zoneinfo *zp;
|
|
|
|
zp = zh.opaque;
|
|
INSIST(zp != NULL);
|
|
|
|
/* Fail if filename already set for this zone */
|
|
if (zp->z_source != NULL)
|
|
return (0);
|
|
zp->z_source = filename;
|
|
return (1);
|
|
}
|
|
|
|
int
|
|
set_zone_checknames(zone_config zh, enum severity s) {
|
|
struct zoneinfo *zp;
|
|
|
|
zp = zh.opaque;
|
|
INSIST(zp != NULL);
|
|
|
|
/* Fail if checknames already set for this zone */
|
|
if (zp->z_checknames != not_set)
|
|
return (0);
|
|
zp->z_checknames = s;
|
|
return (1);
|
|
}
|
|
|
|
int
|
|
set_zone_ixfr_file(zone_config zh, char *filename) {
|
|
struct zoneinfo *zp;
|
|
|
|
zp = zh.opaque;
|
|
INSIST(zp != NULL);
|
|
|
|
/* Fail if filename already set for this zone */
|
|
if (zp->z_ixfr_base != NULL)
|
|
return (0);
|
|
zp->z_ixfr_base = filename;
|
|
if (zp->z_ixfr_tmp == NULL) {
|
|
int len = strlen(zp->z_ixfr_base) + (sizeof ".tmp");
|
|
char *str = (char *) memget(len);
|
|
|
|
sprintf(str, "%s.tmp", zp->z_ixfr_base);
|
|
zp->z_ixfr_tmp = savestr(str, 1);
|
|
memput(str, len);
|
|
}
|
|
|
|
return (1);
|
|
}
|
|
|
|
int
|
|
set_zone_ixfr_tmp(zone_config zh, char *filename) {
|
|
struct zoneinfo *zp;
|
|
|
|
zp = zh.opaque;
|
|
INSIST(zp != NULL);
|
|
|
|
/* Fail if filename already set for this zone */
|
|
if (zp->z_ixfr_tmp != NULL)
|
|
return (0);
|
|
zp->z_ixfr_tmp = filename;
|
|
return (1);
|
|
}
|
|
|
|
int
|
|
set_zone_dialup(zone_config zh, int value) {
|
|
struct zoneinfo *zp;
|
|
|
|
zp = zh.opaque;
|
|
INSIST(zp != NULL);
|
|
|
|
if (value) {
|
|
zp->z_dialup = zdialup_yes;
|
|
#ifdef BIND_NOTIFY
|
|
zp->z_notify = znotify_yes;
|
|
#endif
|
|
} else
|
|
zp->z_dialup = zdialup_no;
|
|
|
|
return (1);
|
|
}
|
|
|
|
int
|
|
set_zone_notify(zone_config zh, int value) {
|
|
#ifdef BIND_NOTIFY
|
|
struct zoneinfo *zp;
|
|
|
|
zp = zh.opaque;
|
|
INSIST(zp != NULL);
|
|
|
|
if (value)
|
|
zp->z_notify = znotify_yes;
|
|
else
|
|
zp->z_notify = znotify_no;
|
|
#endif
|
|
return (1);
|
|
}
|
|
|
|
int
|
|
set_zone_maintain_ixfr_base(zone_config zh, int value) {
|
|
struct zoneinfo *zp;
|
|
|
|
zp = zh.opaque;
|
|
INSIST(zp != NULL);
|
|
zp->z_maintain_ixfr_base = value;
|
|
|
|
return (1);
|
|
}
|
|
|
|
int
|
|
set_zone_update_acl(zone_config zh, ip_match_list iml) {
|
|
struct zoneinfo *zp;
|
|
|
|
zp = zh.opaque;
|
|
INSIST(zp != NULL);
|
|
|
|
/* Fail if update_acl already set for this zone */
|
|
if (zp->z_update_acl != NULL)
|
|
return (0);
|
|
zp->z_update_acl = iml;
|
|
#ifdef BIND_UPDATE
|
|
if (!ip_match_is_none(iml))
|
|
zp->z_flags |= Z_DYNAMIC;
|
|
else
|
|
ns_debug(ns_log_config, 3, "update acl is none for '%s'",
|
|
zp->z_origin);
|
|
#endif
|
|
return (1);
|
|
}
|
|
|
|
int
|
|
set_zone_query_acl(zone_config zh, ip_match_list iml) {
|
|
struct zoneinfo *zp;
|
|
|
|
zp = zh.opaque;
|
|
INSIST(zp != NULL);
|
|
|
|
/* Fail if checknames already set for this zone */
|
|
if (zp->z_query_acl != NULL)
|
|
return (0);
|
|
zp->z_query_acl = iml;
|
|
return (1);
|
|
}
|
|
|
|
int
|
|
set_zone_master_port(zone_config zh, u_short port) {
|
|
struct zoneinfo *zp = zh.opaque;
|
|
|
|
zp->z_port = port;
|
|
return (1);
|
|
}
|
|
|
|
int
|
|
set_zone_transfer_source(zone_config zh, struct in_addr ina) {
|
|
struct zoneinfo *zp = zh.opaque;
|
|
|
|
zp->z_axfr_src = ina;
|
|
return (1);
|
|
}
|
|
|
|
int
|
|
set_zone_transfer_acl(zone_config zh, ip_match_list iml) {
|
|
struct zoneinfo *zp;
|
|
|
|
zp = zh.opaque;
|
|
INSIST(zp != NULL);
|
|
|
|
/* Fail if checknames already set for this zone */
|
|
if (zp->z_transfer_acl != NULL)
|
|
return (0);
|
|
zp->z_transfer_acl = iml;
|
|
return (1);
|
|
}
|
|
|
|
int
|
|
set_zone_transfer_time_in(zone_config zh, long max_time) {
|
|
struct zoneinfo *zp;
|
|
|
|
zp = zh.opaque;
|
|
INSIST(zp != NULL);
|
|
|
|
/* Fail if checknames already set for this zone */
|
|
if (zp->z_max_transfer_time_in)
|
|
return (0);
|
|
zp->z_max_transfer_time_in = max_time;
|
|
return (1);
|
|
}
|
|
|
|
int
|
|
set_zone_max_log_size_ixfr(zone_config zh, int size) {
|
|
struct zoneinfo *zp;
|
|
|
|
zp = zh.opaque;
|
|
INSIST(zp != NULL);
|
|
|
|
zp->z_max_log_size_ixfr = size;
|
|
return (0);
|
|
}
|
|
|
|
int
|
|
set_zone_pubkey(zone_config zh, const int flags, const int proto,
|
|
const int alg, const char *str)
|
|
{
|
|
struct zoneinfo *zp;
|
|
|
|
zp = zh.opaque;
|
|
INSIST(zp != NULL);
|
|
|
|
INSIST(zp != NULL && zp->z_origin != NULL);
|
|
return (add_trusted_key(zp->z_origin, flags, proto, alg, str));
|
|
}
|
|
|
|
int
|
|
set_trusted_key(const char *name, const int flags, const int proto,
|
|
const int alg, const char *str) {
|
|
INSIST(name != NULL);
|
|
return (add_trusted_key(name, flags, proto, alg, str));
|
|
}
|
|
|
|
int
|
|
add_zone_master(zone_config zh, struct in_addr address) {
|
|
struct zoneinfo *zp;
|
|
|
|
zp = zh.opaque;
|
|
INSIST(zp != NULL);
|
|
|
|
zp->z_addr[zp->z_addrcnt] = address;
|
|
zp->z_addrcnt++;
|
|
if (zp->z_addrcnt >= NSMAX) {
|
|
ns_warning(ns_log_config, "NSMAX reached for zone '%s'",
|
|
zp->z_origin);
|
|
zp->z_addrcnt = NSMAX - 1;
|
|
}
|
|
return (1);
|
|
}
|
|
|
|
int
|
|
add_zone_notify(zone_config zh, struct in_addr address) {
|
|
#ifdef BIND_NOTIFY
|
|
struct zoneinfo *zp;
|
|
int i;
|
|
|
|
zp = zh.opaque;
|
|
INSIST(zp != NULL);
|
|
|
|
/* Check for duplicates. */
|
|
|
|
for (i = 0; i < zp->z_notify_count; i++) {
|
|
if (memcmp(zp->z_also_notify + i,
|
|
&address, sizeof address) == 0) {
|
|
ns_warning(ns_log_config,
|
|
"duplicate also-notify address ignored [%s] for zone '%s'",
|
|
inet_ntoa(address), zp->z_origin);
|
|
return (1);
|
|
}
|
|
}
|
|
i = 0;
|
|
|
|
if (zp->z_also_notify == NULL) {
|
|
zp->z_also_notify = memget(sizeof *zp->z_also_notify);
|
|
if (zp->z_also_notify == NULL)
|
|
i = 1;
|
|
} else {
|
|
register size_t size;
|
|
register struct in_addr *an_tmp;
|
|
size = zp->z_notify_count * sizeof *zp->z_also_notify;
|
|
an_tmp = memget(size + sizeof *zp->z_also_notify);
|
|
if (an_tmp == NULL) {
|
|
i = 1;
|
|
} else {
|
|
memcpy(an_tmp, zp->z_also_notify, size);
|
|
memput(zp->z_also_notify, size);
|
|
zp->z_also_notify = an_tmp;
|
|
}
|
|
}
|
|
if (i == 0) {
|
|
zp->z_also_notify[zp->z_notify_count] = address;
|
|
zp->z_notify_count++;
|
|
} else {
|
|
ns_warning(ns_log_config, "also-notify add failed (memget) [%s] for zone '%s'",
|
|
inet_ntoa(address), zp->z_origin);
|
|
}
|
|
#endif
|
|
return (1);
|
|
}
|
|
|
|
/* Options */
|
|
|
|
options
|
|
new_options() {
|
|
options op;
|
|
|
|
op = (options)memget(sizeof (struct options));
|
|
if (op == NULL)
|
|
panic("memget failed in new_options()", NULL);
|
|
|
|
op->version = savestr(ShortVersion, 1);
|
|
op->directory = savestr(".", 1);
|
|
op->pid_filename = savestr(_PATH_PIDFILE, 1);
|
|
op->named_xfer = savestr(_PATH_XFER, 1);
|
|
op->dump_filename = savestr(_PATH_DUMPFILE, 1);
|
|
op->stats_filename = savestr(_PATH_STATS, 1);
|
|
op->memstats_filename = savestr(_PATH_MEMSTATS, 1);
|
|
op->flags = DEFAULT_OPTION_FLAGS;
|
|
op->transfers_in = DEFAULT_XFERS_RUNNING;
|
|
op->transfers_per_ns = DEFAULT_XFERS_PER_NS;
|
|
op->transfers_out = 0;
|
|
op->serial_queries = MAXQSERIAL;
|
|
op->transfer_format = axfr_one_answer;
|
|
op->max_transfer_time_in = MAX_XFER_TIME;
|
|
memset(&op->query_source, 0, sizeof op->query_source);
|
|
op->query_source.sin_family = AF_INET;
|
|
op->query_source.sin_addr.s_addr = htonl(INADDR_ANY);
|
|
op->query_source.sin_port = htons(0); /* INPORT_ANY */
|
|
op->axfr_src.s_addr = 0;
|
|
#ifdef BIND_NOTIFY
|
|
op->notify_count = 0;
|
|
op->also_notify = NULL;
|
|
#endif
|
|
op->blackhole_acl = NULL;
|
|
op->query_acl = NULL;
|
|
op->transfer_acl = NULL;
|
|
op->recursion_acl = NULL;
|
|
op->sortlist = NULL;
|
|
op->topology = NULL;
|
|
op->data_size = 0UL; /* use system default */
|
|
op->stack_size = 0UL; /* use system default */
|
|
op->core_size = 0UL; /* use system default */
|
|
op->files = ULONG_MAX; /* unlimited */
|
|
op->check_names[primary_trans] = fail;
|
|
op->check_names[secondary_trans] = warn;
|
|
op->check_names[response_trans] = ignore;
|
|
op->listen_list = NULL;
|
|
op->fwdtab = NULL;
|
|
/* XXX init forwarding */
|
|
op->clean_interval = 3600;
|
|
op->interface_interval = 3600;
|
|
op->stats_interval = 3600;
|
|
op->ordering = NULL;
|
|
op->max_ncache_ttl = DEFAULT_MAX_NCACHE_TTL;
|
|
op->max_host_stats = 0;
|
|
op->lame_ttl = NTTL;
|
|
op->heartbeat_interval = 3600;
|
|
op->max_log_size_ixfr = 20;
|
|
op->minroots = MINROOTS;
|
|
return (op);
|
|
}
|
|
|
|
void
|
|
free_options(options op) {
|
|
INSIST(op != NULL);
|
|
|
|
if (op->version)
|
|
freestr(op->version);
|
|
if (op->directory)
|
|
freestr(op->directory);
|
|
if (op->pid_filename)
|
|
freestr(op->pid_filename);
|
|
if (op->named_xfer)
|
|
freestr(op->named_xfer);
|
|
if (op->dump_filename)
|
|
freestr(op->dump_filename);
|
|
if (op->stats_filename)
|
|
freestr(op->stats_filename);
|
|
if (op->memstats_filename)
|
|
freestr(op->memstats_filename);
|
|
#ifdef BIND_NOTIFY
|
|
if (op->also_notify)
|
|
free_also_notify(op);
|
|
#endif
|
|
if (op->blackhole_acl)
|
|
free_ip_match_list(op->blackhole_acl);
|
|
if (op->query_acl)
|
|
free_ip_match_list(op->query_acl);
|
|
if (op->recursion_acl)
|
|
free_ip_match_list(op->recursion_acl);
|
|
if (op->transfer_acl)
|
|
free_ip_match_list(op->transfer_acl);
|
|
if (op->sortlist)
|
|
free_ip_match_list(op->sortlist);
|
|
if (op->ordering)
|
|
free_rrset_order_list(op->ordering);
|
|
if (op->topology)
|
|
free_ip_match_list(op->topology);
|
|
if (op->listen_list)
|
|
free_listen_info_list(op->listen_list);
|
|
if (op->fwdtab)
|
|
free_forwarders(op->fwdtab);
|
|
memput(op, sizeof *op);
|
|
}
|
|
|
|
static void
|
|
set_boolean_option(u_int *op_flags, int bool_opt, int value) {
|
|
INSIST(op_flags != NULL);
|
|
|
|
switch (bool_opt) {
|
|
#ifdef HITCOUNTS
|
|
case OPTION_HITCOUNT:
|
|
#endif /* HITCOUNTS */
|
|
case OPTION_NORECURSE:
|
|
case OPTION_NOFETCHGLUE:
|
|
case OPTION_FORWARD_ONLY:
|
|
case OPTION_FAKE_IQUERY:
|
|
case OPTION_NONOTIFY:
|
|
case OPTION_NONAUTH_NXDOMAIN:
|
|
case OPTION_MULTIPLE_CNAMES:
|
|
case OPTION_USE_IXFR:
|
|
case OPTION_MAINTAIN_IXFR_BASE:
|
|
case OPTION_HOSTSTATS:
|
|
case OPTION_DEALLOC_ON_EXIT:
|
|
case OPTION_USE_ID_POOL:
|
|
case OPTION_NORFC2308_TYPE1:
|
|
case OPTION_NODIALUP:
|
|
case OPTION_TREAT_CR_AS_SPACE:
|
|
if (value)
|
|
*op_flags |= bool_opt;
|
|
else
|
|
*op_flags &= ~bool_opt;
|
|
break;
|
|
default:
|
|
panic("unexpected option in set_boolean_option", NULL);
|
|
}
|
|
}
|
|
|
|
void
|
|
set_global_boolean_option(options op, int bool_opt, int value) {
|
|
|
|
INSIST(op != NULL);
|
|
|
|
set_boolean_option(&op->flags, bool_opt, value);
|
|
}
|
|
|
|
void
|
|
set_zone_boolean_option(zone_config zh, int bool_opt, int value) {
|
|
struct zoneinfo *zp;
|
|
|
|
zp = zh.opaque;
|
|
INSIST(zp != NULL);
|
|
|
|
set_boolean_option(&zp->z_options, bool_opt, value);
|
|
|
|
/* Flag that zone option overrides corresponding global option */
|
|
zp->z_optset |= bool_opt;
|
|
}
|
|
|
|
#ifdef HAVE_GETRUSAGE
|
|
enum limit { Datasize, Stacksize, Coresize, Files };
|
|
|
|
static struct rlimit initial_data_size;
|
|
static struct rlimit initial_stack_size;
|
|
static struct rlimit initial_core_size;
|
|
static struct rlimit initial_num_files;
|
|
|
|
static void
|
|
get_initial_limits() {
|
|
int fdlimit = evHighestFD(ev) + 1;
|
|
|
|
# ifdef RLIMIT_DATA
|
|
if (getrlimit(RLIMIT_DATA, &initial_data_size) < 0)
|
|
ns_warning(ns_log_config, "getrlimit(DATA): %s",
|
|
strerror(errno));
|
|
# endif
|
|
# ifdef RLIMIT_STACK
|
|
if (getrlimit(RLIMIT_STACK, &initial_stack_size) < 0)
|
|
ns_warning(ns_log_config, "getrlimit(STACK): %s",
|
|
strerror(errno));
|
|
# endif
|
|
# ifdef RLIMIT_CORE
|
|
if (getrlimit(RLIMIT_CORE, &initial_core_size) < 0)
|
|
ns_warning(ns_log_config, "getrlimit(CORE): %s",
|
|
strerror(errno));
|
|
# endif
|
|
# ifdef RLIMIT_NOFILE
|
|
if (getrlimit(RLIMIT_NOFILE, &initial_num_files) < 0)
|
|
ns_warning(ns_log_config, "getrlimit(NOFILE): %s",
|
|
strerror(errno));
|
|
else if (initial_num_files.rlim_cur > fdlimit) {
|
|
initial_num_files.rlim_cur = fdlimit;
|
|
if (initial_num_files.rlim_cur > initial_num_files.rlim_max)
|
|
initial_num_files.rlim_max = fdlimit;
|
|
if (setrlimit(RLIMIT_NOFILE, &initial_num_files) < 0) {
|
|
ns_warning(ns_log_config, "setrlimit(files): %s",
|
|
strerror(errno));
|
|
} else {
|
|
ns_warning(ns_log_config,
|
|
"limit files set to fdlimit (%d)",
|
|
fdlimit);
|
|
}
|
|
}
|
|
# endif
|
|
}
|
|
|
|
static void
|
|
ns_rlimit(enum limit limit, u_long limit_value) {
|
|
struct rlimit limits, old_limits;
|
|
int rlimit = -1;
|
|
int fdlimit = evHighestFD(ev) + 1;
|
|
char *name;
|
|
rlimit_type value;
|
|
|
|
if (limit_value == ULONG_MAX) {
|
|
#ifndef RLIMIT_FILE_INFINITY
|
|
if (limit == Files)
|
|
value = MIN((rlimit_type)evHighestFD(ev) + 1,
|
|
initial_num_files.rlim_max);
|
|
else
|
|
#endif
|
|
value = (rlimit_type)RLIM_INFINITY;
|
|
} else
|
|
value = (rlimit_type)limit_value;
|
|
|
|
limits.rlim_cur = limits.rlim_max = value;
|
|
switch (limit) {
|
|
case Datasize:
|
|
#ifdef RLIMIT_DATA
|
|
rlimit = RLIMIT_DATA;
|
|
#endif
|
|
name = "max data size";
|
|
if (value == 0)
|
|
limits = initial_data_size;
|
|
break;
|
|
case Stacksize:
|
|
#ifdef RLIMIT_STACK
|
|
rlimit = RLIMIT_STACK;
|
|
#endif
|
|
name = "max stack size";
|
|
if (value == 0)
|
|
limits = initial_stack_size;
|
|
break;
|
|
case Coresize:
|
|
#ifdef RLIMIT_CORE
|
|
rlimit = RLIMIT_CORE;
|
|
#endif
|
|
name = "max core size";
|
|
if (value == 0)
|
|
limits = initial_core_size;
|
|
break;
|
|
case Files:
|
|
#ifdef RLIMIT_NOFILE
|
|
rlimit = RLIMIT_NOFILE;
|
|
#endif
|
|
name = "max number of open files";
|
|
if (value == 0)
|
|
limits = initial_num_files;
|
|
if (value > fdlimit)
|
|
limits.rlim_cur = limits.rlim_max = value = fdlimit;
|
|
break;
|
|
default:
|
|
name = NULL; /* Make gcc happy. */
|
|
panic("impossible condition in ns_rlimit()", NULL);
|
|
}
|
|
if (rlimit == -1) {
|
|
ns_warning(ns_log_config,
|
|
"limit \"%s\" not supported on this system - ignored",
|
|
name);
|
|
return;
|
|
}
|
|
if (getrlimit(rlimit, &old_limits) < 0) {
|
|
ns_warning(ns_log_config, "getrlimit(%s): %s", name,
|
|
strerror(errno));
|
|
}
|
|
if (user_id != 0 && limits.rlim_max == RLIM_INFINITY)
|
|
limits.rlim_cur = limits.rlim_max = old_limits.rlim_max;
|
|
if (setrlimit(rlimit, &limits) < 0) {
|
|
ns_warning(ns_log_config, "setrlimit(%s): %s", name,
|
|
strerror(errno));
|
|
return;
|
|
} else {
|
|
if (value == 0)
|
|
ns_debug(ns_log_config, 3, "%s is default", name);
|
|
else if (value == RLIM_INFINITY)
|
|
ns_debug(ns_log_config, 3, "%s is unlimited", name);
|
|
else
|
|
#ifdef RLIMIT_LONGLONG
|
|
ns_debug(ns_log_config, 3, "%s is %llu", name,
|
|
(unsigned long long)value);
|
|
#else
|
|
ns_debug(ns_log_config, 3, "%s is %lu", name, value);
|
|
#endif
|
|
}
|
|
}
|
|
#endif /* HAVE_GETRUSAGE */
|
|
|
|
listen_info_list
|
|
new_listen_info_list() {
|
|
listen_info_list ll;
|
|
|
|
ll = (listen_info_list)memget(sizeof (struct listen_info_list));
|
|
if (ll == NULL)
|
|
panic("memget failed in new_listen_info_list()", NULL);
|
|
ll->first = NULL;
|
|
ll->last = NULL;
|
|
return (ll);
|
|
}
|
|
|
|
void
|
|
free_listen_info_list(listen_info_list ll) {
|
|
listen_info li, next_li;
|
|
|
|
INSIST(ll != NULL);
|
|
for (li = ll->first; li != NULL; li = next_li) {
|
|
next_li = li->next;
|
|
free_ip_match_list(li->list);
|
|
memput(li, sizeof *li);
|
|
}
|
|
memput(ll, sizeof *ll);
|
|
}
|
|
|
|
void
|
|
add_listen_on(options op, u_short port, ip_match_list iml) {
|
|
listen_info_list ll;
|
|
listen_info ni;
|
|
|
|
INSIST(op != NULL);
|
|
|
|
if (op->listen_list == NULL)
|
|
op->listen_list = new_listen_info_list();
|
|
ll = op->listen_list;
|
|
ni = (listen_info)memget(sizeof (struct listen_info));
|
|
if (ni == NULL)
|
|
panic("memget failed in add_listen_on", NULL);
|
|
ni->port = port;
|
|
ni->list = iml;
|
|
ni->next = NULL;
|
|
if (ll->last != NULL)
|
|
ll->last->next = ni;
|
|
ll->last = ni;
|
|
if (ll->first == NULL)
|
|
ll->first = ni;
|
|
}
|
|
|
|
FILE *
|
|
write_open(char *filename) {
|
|
FILE *stream;
|
|
int fd;
|
|
struct stat sb;
|
|
int regular;
|
|
|
|
if (stat(filename, &sb) < 0) {
|
|
if (errno != ENOENT) {
|
|
ns_error(ns_log_os,
|
|
"write_open: stat of %s failed: %s",
|
|
filename, strerror(errno));
|
|
return (NULL);
|
|
}
|
|
regular = 1;
|
|
} else
|
|
regular = (sb.st_mode & S_IFREG);
|
|
|
|
if (!regular) {
|
|
ns_error(ns_log_os, "write_open: %s isn't a regular file",
|
|
filename);
|
|
return (NULL);
|
|
}
|
|
|
|
(void)unlink(filename);
|
|
fd = open(filename, O_WRONLY|O_CREAT|O_EXCL,
|
|
S_IRUSR|S_IWUSR|S_IRGRP|S_IWGRP|S_IROTH|S_IWOTH);
|
|
if (fd < 0)
|
|
return (NULL);
|
|
(void) fchown(fd, user_id, group_id);
|
|
stream = fdopen(fd, "w");
|
|
if (stream == NULL)
|
|
(void)close(fd);
|
|
return (stream);
|
|
}
|
|
|
|
void
|
|
update_pid_file() {
|
|
FILE *fp;
|
|
|
|
REQUIRE(server_options != NULL);
|
|
REQUIRE(server_options->pid_filename != NULL);
|
|
|
|
/* XXX */ ns_debug(ns_log_default, 1, "update_pid_file()");
|
|
if (current_pid_filename != NULL) {
|
|
(void)unlink(current_pid_filename);
|
|
freestr(current_pid_filename);
|
|
current_pid_filename = NULL;
|
|
}
|
|
current_pid_filename = savestr(server_options->pid_filename, 0);
|
|
if (current_pid_filename == NULL) {
|
|
ns_error(ns_log_config,
|
|
"savestr() failed in update_pid_file()");
|
|
return;
|
|
}
|
|
fp = write_open(current_pid_filename);
|
|
if (fp != NULL) {
|
|
(void) fprintf(fp, "%ld\n", (long)getpid());
|
|
(void) fclose(fp);
|
|
} else
|
|
ns_error(ns_log_config, "couldn't create pid file '%s'",
|
|
server_options->pid_filename);
|
|
}
|
|
|
|
/*
|
|
* XXX This function will eventually be public and will be relocated to
|
|
* the UNIX OS support library.
|
|
*/
|
|
|
|
static int
|
|
os_change_directory(const char *name) {
|
|
struct stat sb;
|
|
|
|
if (name == NULL ||
|
|
*name == '\0') {
|
|
errno = EINVAL;
|
|
return (0);
|
|
}
|
|
|
|
if (chdir(name) < 0)
|
|
return (0);
|
|
|
|
if (stat(name, &sb) < 0) {
|
|
ns_error(ns_log_os, "stat(%s) failed: %s", name,
|
|
strerror(errno));
|
|
return (1);
|
|
}
|
|
if (sb.st_mode & S_IWOTH)
|
|
ns_warning(ns_log_os, "directory %s is world-writable", name);
|
|
|
|
return (1);
|
|
}
|
|
|
|
static void
|
|
periodic_getnetconf(evContext ctx, void *uap, struct timespec due,
|
|
struct timespec inter)
|
|
{
|
|
getnetconf(1);
|
|
}
|
|
|
|
static int clean_interval = 0;
|
|
static int interface_interval = 0;
|
|
static int stats_interval = 0;
|
|
static int heartbeat_interval = 0;
|
|
|
|
static void
|
|
set_interval_timer(int which_timer, int interval) {
|
|
evTimerID *tid = NULL;
|
|
evTimerFunc func = NULL;
|
|
int changed = 0;
|
|
|
|
switch (which_timer) {
|
|
case CLEAN_TIMER:
|
|
if (clean_interval != interval)
|
|
changed = 1;
|
|
clean_interval = interval;
|
|
tid = &clean_timer;
|
|
func = ns_cleancache;
|
|
break;
|
|
case INTERFACE_TIMER:
|
|
if (interface_interval != interval)
|
|
changed = 1;
|
|
interface_interval = interval;
|
|
tid = &interface_timer;
|
|
func = periodic_getnetconf;
|
|
break;
|
|
case STATS_TIMER:
|
|
if (stats_interval != interval)
|
|
changed = 1;
|
|
stats_interval = interval;
|
|
tid = &stats_timer;
|
|
func = ns_logstats;
|
|
break;
|
|
case HEARTBEAT_TIMER:
|
|
if (heartbeat_interval != interval)
|
|
changed = 1;
|
|
heartbeat_interval = interval;
|
|
tid = &heartbeat_timer;
|
|
func = ns_heartbeat;
|
|
break;
|
|
default:
|
|
ns_panic(ns_log_config, 1,
|
|
"set_interval_timer: unknown timer %d", which_timer);
|
|
}
|
|
if ((active_timers & which_timer) != 0) {
|
|
if (interval > 0) {
|
|
if (changed &&
|
|
evResetTimer(ev, *tid, func, NULL,
|
|
evAddTime(evNowTime(),
|
|
evConsTime(interval, 0)),
|
|
evConsTime(interval, 0)) < 0)
|
|
ns_error(ns_log_config,
|
|
"evResetTimer %d interval %d failed: %s",
|
|
which_timer, interval,
|
|
strerror(errno));
|
|
} else {
|
|
if (evClearTimer(ev, *tid) < 0)
|
|
ns_error(ns_log_config,
|
|
"evClearTimer %d failed: %s",
|
|
which_timer, strerror(errno));
|
|
else
|
|
active_timers &= ~which_timer;
|
|
}
|
|
} else if (interval > 0) {
|
|
if (evSetTimer(ev, func, NULL,
|
|
evAddTime(evNowTime(),
|
|
evConsTime(interval, 0)),
|
|
evConsTime(interval, 0), tid) < 0)
|
|
ns_error(ns_log_config,
|
|
"evSetTimer %d interval %d failed: %s",
|
|
which_timer, interval, strerror(errno));
|
|
else
|
|
active_timers |= which_timer;
|
|
}
|
|
}
|
|
|
|
/*
|
|
* Set all named global options based on the global options structure
|
|
* generated by the parser.
|
|
*/
|
|
void
|
|
set_options(options op, int is_default) {
|
|
INSIST(op != NULL);
|
|
|
|
if (op->listen_list == NULL) {
|
|
ip_match_list iml;
|
|
ip_match_element ime;
|
|
struct in_addr address;
|
|
|
|
op->listen_list = new_listen_info_list();
|
|
|
|
address.s_addr = htonl(INADDR_ANY);
|
|
iml = new_ip_match_list();
|
|
ime = new_ip_match_pattern(address, 0);
|
|
add_to_ip_match_list(iml, ime);
|
|
add_listen_on(op, htons(NS_DEFAULTPORT), iml);
|
|
}
|
|
if (op->topology == NULL) {
|
|
ip_match_list iml;
|
|
ip_match_element ime;
|
|
|
|
/* default topology is { localhost; localnets; } */
|
|
iml = new_ip_match_list();
|
|
ime = new_ip_match_localhost();
|
|
add_to_ip_match_list(iml, ime);
|
|
ime = new_ip_match_localnets();
|
|
add_to_ip_match_list(iml, ime);
|
|
op->topology = iml;
|
|
}
|
|
if (server_options != NULL)
|
|
free_options(server_options);
|
|
server_options = op;
|
|
|
|
/* XXX should validate pid filename */
|
|
INSIST(op->pid_filename != NULL);
|
|
|
|
if (op->directory && !os_change_directory(op->directory))
|
|
ns_panic(ns_log_config, 0, "can't change directory to %s: %s",
|
|
op->directory, strerror(errno));
|
|
|
|
/* XXX currently a value of 0 means "use default"; it would be
|
|
better if the options block had a "attributes updated" vector
|
|
(like the way X deals with GC updates) */
|
|
|
|
if (!op->transfers_in)
|
|
op->transfers_in = DEFAULT_XFERS_RUNNING;
|
|
else if (op->transfers_in > MAX_XFERS_RUNNING) {
|
|
ns_warning(ns_log_config,
|
|
"the maximum number of concurrent inbound transfers is %d",
|
|
MAX_XFERS_RUNNING);
|
|
op->transfers_in = MAX_XFERS_RUNNING;
|
|
}
|
|
|
|
if (!op->transfers_per_ns)
|
|
op->transfers_per_ns = DEFAULT_XFERS_PER_NS;
|
|
|
|
if (!op->max_transfer_time_in)
|
|
op->max_transfer_time_in = MAX_XFER_TIME;
|
|
|
|
/* XXX currently transfers_out is not used */
|
|
|
|
if (!op->max_ncache_ttl)
|
|
op->max_ncache_ttl = DEFAULT_MAX_NCACHE_TTL;
|
|
else if (op->max_ncache_ttl > max_cache_ttl)
|
|
op->max_ncache_ttl = max_cache_ttl;
|
|
|
|
if (op->lame_ttl > (3 * NTTL))
|
|
op->lame_ttl = 3 * NTTL;
|
|
|
|
/*
|
|
* Limits
|
|
*/
|
|
|
|
#ifdef HAVE_GETRUSAGE
|
|
ns_rlimit(Datasize, op->data_size);
|
|
ns_rlimit(Stacksize, op->stack_size);
|
|
ns_rlimit(Coresize, op->core_size);
|
|
ns_rlimit(Files, op->files);
|
|
#else
|
|
ns_info(ns_log_config, "cannot set resource limits on this system");
|
|
#endif
|
|
|
|
/*
|
|
* Timers
|
|
*/
|
|
set_interval_timer(CLEAN_TIMER, server_options->clean_interval);
|
|
set_interval_timer(INTERFACE_TIMER,
|
|
server_options->interface_interval);
|
|
set_interval_timer(STATS_TIMER, server_options->stats_interval);
|
|
set_interval_timer(HEARTBEAT_TIMER,
|
|
server_options->heartbeat_interval);
|
|
|
|
options_installed = 1;
|
|
default_options_installed = is_default;
|
|
}
|
|
|
|
void
|
|
use_default_options() {
|
|
set_options(new_options(), 1);
|
|
}
|
|
|
|
/*
|
|
* rrset order types
|
|
*/
|
|
static struct res_sym order_table [] = {
|
|
{ unknown_order, " unknown " }, /* can't match */
|
|
{ fixed_order, "fixed" },
|
|
{ cyclic_order, "cyclic" },
|
|
{ random_order, "random" },
|
|
{ unknown_order, NULL }
|
|
};
|
|
|
|
/*
|
|
* Return the print name of the ordering value.
|
|
*/
|
|
const char *
|
|
p_order(int order) {
|
|
return (sym_ntos(order_table, order, (int *)0));
|
|
}
|
|
|
|
/*
|
|
* Lookup the ordering by name and return the matching enum value.
|
|
*/
|
|
enum ordering
|
|
lookup_ordering(const char *name) {
|
|
int i;
|
|
|
|
for (i = 0; order_table[i].name != NULL; i++)
|
|
if (strcasecmp(name,order_table[i].name) == 0)
|
|
return ((enum ordering)order_table[i].number);
|
|
return (unknown_order);
|
|
}
|
|
|
|
/*
|
|
* rrset-order Lists
|
|
*/
|
|
rrset_order_list
|
|
new_rrset_order_list() {
|
|
rrset_order_list rol ;
|
|
|
|
rol = (rrset_order_list)memget(sizeof (struct rrset_order_list));
|
|
if (rol == NULL)
|
|
panic("memget failed in new_rrset_order_list", NULL);
|
|
rol->first = NULL;
|
|
rol->last = NULL;
|
|
|
|
return (rol);
|
|
}
|
|
|
|
void
|
|
free_rrset_order_list(rrset_order_list rol) {
|
|
rrset_order_element roe, next_element;
|
|
|
|
for (roe = rol->first; roe != NULL; roe = next_element) {
|
|
next_element = roe->next;
|
|
freestr(roe->name);
|
|
memput(roe, sizeof (*roe));
|
|
}
|
|
memput(rol, sizeof (*rol));
|
|
}
|
|
|
|
void
|
|
add_to_rrset_order_list(rrset_order_list rol, rrset_order_element roe) {
|
|
INSIST(rol != NULL);
|
|
INSIST(roe != NULL);
|
|
|
|
if (rol->last != NULL)
|
|
rol->last->next = roe;
|
|
roe->next = NULL;
|
|
rol->last = roe;
|
|
if (rol->first == NULL)
|
|
rol->first = roe;
|
|
}
|
|
|
|
/* XXX this isn't being used yet, but it probably should be. Where? */
|
|
void
|
|
dprint_rrset_order_list(int category, rrset_order_list rol, int indent,
|
|
char *allow, char *deny) {
|
|
rrset_order_element roe ;
|
|
char spaces[40+1];
|
|
|
|
INSIST(rol != NULL);
|
|
|
|
if (indent > 40)
|
|
indent = 40;
|
|
if (indent)
|
|
memset(spaces, ' ', indent);
|
|
spaces[indent] = '\0';
|
|
|
|
for (roe = rol->first; roe != NULL; roe = roe->next) {
|
|
ns_debug(category, 7, "%sclass %s type %s name %s order %s",
|
|
spaces, p_class(roe->class), p_type(roe->type),
|
|
roe->name, p_order(roe->order));
|
|
}
|
|
}
|
|
|
|
|
|
rrset_order_element
|
|
new_rrset_order_element(int class, int type, char *name, enum ordering order)
|
|
{
|
|
rrset_order_element roe;
|
|
int i ;
|
|
|
|
roe = (rrset_order_element)memget(sizeof (struct rrset_order_element));
|
|
if (roe == NULL)
|
|
panic("memget failed in new_rrset_order_element", NULL);
|
|
roe->class = class ;
|
|
roe->type = type ;
|
|
roe->name = name;
|
|
roe->order = order;
|
|
|
|
i = strlen(roe->name) - 1;
|
|
INSIST (i >= 0);
|
|
if (roe->name[i - 1] == '.') {
|
|
/* We compare from right to left so we don't need a dot on
|
|
the end. */
|
|
roe->name[i - 1] = '\0' ;
|
|
}
|
|
|
|
return roe ;
|
|
}
|
|
|
|
|
|
/*
|
|
* IP Matching Lists
|
|
*/
|
|
|
|
ip_match_list
|
|
new_ip_match_list() {
|
|
ip_match_list iml;
|
|
|
|
iml = (ip_match_list)memget(sizeof (struct ip_match_list));
|
|
if (iml == NULL)
|
|
panic("memget failed in new_ip_match_list", NULL);
|
|
iml->first = NULL;
|
|
iml->last = NULL;
|
|
return (iml);
|
|
}
|
|
|
|
void
|
|
free_ip_match_list(ip_match_list iml) {
|
|
ip_match_element ime, next_element;
|
|
|
|
for (ime = iml->first; ime != NULL; ime = next_element) {
|
|
next_element = ime->next;
|
|
memput(ime, sizeof *ime);
|
|
}
|
|
memput(iml, sizeof *iml);
|
|
}
|
|
|
|
ip_match_element
|
|
new_ip_match_pattern(struct in_addr address, u_int mask_bits) {
|
|
ip_match_element ime;
|
|
u_int32_t mask;
|
|
|
|
ime = (ip_match_element)memget(sizeof (struct ip_match_element));
|
|
if (ime == NULL)
|
|
panic("memget failed in new_ip_match_pattern", NULL);
|
|
ime->type = ip_match_pattern;
|
|
ime->flags = 0;
|
|
ime->u.direct.address = address;
|
|
if (mask_bits == 0)
|
|
/* can't shift >= the size of a type in bits, so
|
|
we deal with an empty mask here */
|
|
mask = 0;
|
|
else {
|
|
/* set the 'mask_bits' most significant bits */
|
|
mask = 0xffffffffU;
|
|
mask >>= (32 - mask_bits);
|
|
mask <<= (32 - mask_bits);
|
|
}
|
|
mask = ntohl(mask);
|
|
ime->u.direct.mask.s_addr = mask;
|
|
ime->next = NULL;
|
|
if (!ina_onnet(ime->u.direct.address, ime->u.direct.address,
|
|
ime->u.direct.mask)) {
|
|
memput(ime, sizeof *ime);
|
|
ime = NULL;
|
|
}
|
|
return (ime);
|
|
}
|
|
|
|
ip_match_element
|
|
new_ip_match_mask(struct in_addr address, struct in_addr mask) {
|
|
ip_match_element ime;
|
|
|
|
ime = (ip_match_element)memget(sizeof (struct ip_match_element));
|
|
if (ime == NULL)
|
|
panic("memget failed in new_ip_match_pattern", NULL);
|
|
ime->type = ip_match_pattern;
|
|
ime->flags = 0;
|
|
ime->u.direct.address = address;
|
|
ime->u.direct.mask = mask;
|
|
ime->next = NULL;
|
|
if (!ina_onnet(ime->u.direct.address, ime->u.direct.address,
|
|
ime->u.direct.mask)) {
|
|
memput(ime, sizeof *ime);
|
|
ime = NULL;
|
|
}
|
|
return (ime);
|
|
}
|
|
|
|
ip_match_element
|
|
new_ip_match_indirect(ip_match_list iml) {
|
|
ip_match_element ime;
|
|
|
|
INSIST(iml != NULL);
|
|
|
|
ime = (ip_match_element)memget(sizeof (struct ip_match_element));
|
|
if (ime == NULL)
|
|
panic("memget failed in new_ip_match_indirect", NULL);
|
|
ime->type = ip_match_indirect;
|
|
ime->flags = 0;
|
|
ime->u.indirect.list = iml;
|
|
ime->next = NULL;
|
|
return (ime);
|
|
}
|
|
|
|
ip_match_element
|
|
new_ip_match_key(DST_KEY *dst_key) {
|
|
ip_match_element ime;
|
|
|
|
ime = (ip_match_element)memget(sizeof (struct ip_match_element));
|
|
if (ime == NULL)
|
|
panic("memget failed in new_ip_match_key", NULL);
|
|
ime->type = ip_match_key;
|
|
ime->flags = 0;
|
|
ime->u.key.key = dst_key;
|
|
return (ime);
|
|
}
|
|
|
|
ip_match_element
|
|
new_ip_match_localhost() {
|
|
ip_match_element ime;
|
|
|
|
ime = (ip_match_element)memget(sizeof (struct ip_match_element));
|
|
if (ime == NULL)
|
|
panic("memget failed in new_ip_match_localhost", NULL);
|
|
ime->type = ip_match_localhost;
|
|
ime->flags = 0;
|
|
ime->u.indirect.list = NULL;
|
|
ime->next = NULL;
|
|
return (ime);
|
|
}
|
|
|
|
ip_match_element
|
|
new_ip_match_localnets() {
|
|
ip_match_element ime;
|
|
|
|
ime = (ip_match_element)memget(sizeof (struct ip_match_element));
|
|
if (ime == NULL)
|
|
panic("memget failed in new_ip_match_localnets", NULL);
|
|
ime->type = ip_match_localnets;
|
|
ime->flags = 0;
|
|
ime->u.indirect.list = NULL;
|
|
ime->next = NULL;
|
|
return (ime);
|
|
}
|
|
|
|
void
|
|
ip_match_negate(ip_match_element ime) {
|
|
if (ime->flags & IP_MATCH_NEGATE)
|
|
ime->flags &= ~IP_MATCH_NEGATE;
|
|
else
|
|
ime->flags |= IP_MATCH_NEGATE;
|
|
}
|
|
|
|
void
|
|
add_to_ip_match_list(ip_match_list iml, ip_match_element ime) {
|
|
INSIST(iml != NULL);
|
|
INSIST(ime != NULL);
|
|
|
|
if (iml->last != NULL)
|
|
iml->last->next = ime;
|
|
ime->next = NULL;
|
|
iml->last = ime;
|
|
if (iml->first == NULL)
|
|
iml->first = ime;
|
|
}
|
|
|
|
void
|
|
dprint_ip_match_list(int category, ip_match_list iml, int indent,
|
|
char *allow, char *deny) {
|
|
ip_match_element ime;
|
|
char spaces[40+1];
|
|
char addr_text[sizeof "255.255.255.255"];
|
|
char mask_text[sizeof "255.255.255.255"];
|
|
|
|
INSIST(iml != NULL);
|
|
|
|
if (indent > 40)
|
|
indent = 40;
|
|
if (indent)
|
|
memset(spaces, ' ', indent);
|
|
spaces[indent] = '\0';
|
|
|
|
for (ime = iml->first; ime != NULL; ime = ime->next) {
|
|
switch (ime->type) {
|
|
case ip_match_pattern:
|
|
memset(addr_text, 0, sizeof addr_text);
|
|
strncpy(addr_text, inet_ntoa(ime->u.direct.address),
|
|
((sizeof addr_text) - 1));
|
|
memset(mask_text, 0, sizeof mask_text);
|
|
strncpy(mask_text, inet_ntoa(ime->u.direct.mask),
|
|
((sizeof mask_text) - 1));
|
|
ns_debug(category, 7, "%s%saddr: %s, mask: %s",
|
|
spaces,
|
|
(ime->flags & IP_MATCH_NEGATE) ? deny : allow,
|
|
addr_text, mask_text);
|
|
break;
|
|
case ip_match_localhost:
|
|
ns_debug(category, 7, "%s%slocalhost", spaces,
|
|
(ime->flags & IP_MATCH_NEGATE) ?
|
|
deny : allow);
|
|
break;
|
|
case ip_match_localnets:
|
|
ns_debug(category, 7, "%s%slocalnets", spaces,
|
|
(ime->flags & IP_MATCH_NEGATE) ?
|
|
deny : allow);
|
|
break;
|
|
case ip_match_indirect:
|
|
ns_debug(category, 7, "%s%sindirect list %p", spaces,
|
|
(ime->flags & IP_MATCH_NEGATE) ? deny : allow,
|
|
ime->u.indirect.list);
|
|
if (ime->u.indirect.list != NULL)
|
|
dprint_ip_match_list(category,
|
|
ime->u.indirect.list,
|
|
indent+2, allow, deny);
|
|
break;
|
|
case ip_match_key:
|
|
ns_debug(category, 7, "%s%skey %s", spaces,
|
|
(ime->flags & IP_MATCH_NEGATE) ? deny : allow,
|
|
ime->u.key.key->dk_key_name);
|
|
break;
|
|
default:
|
|
panic("unexpected ime type in dprint_ip_match_list()",
|
|
NULL);
|
|
}
|
|
}
|
|
}
|
|
|
|
int
|
|
ip_match_addr_or_key(ip_match_list iml, struct in_addr address,
|
|
DST_KEY *key)
|
|
{
|
|
ip_match_element ime;
|
|
int ret;
|
|
int indirect;
|
|
|
|
INSIST(iml != NULL);
|
|
for (ime = iml->first; ime != NULL; ime = ime->next) {
|
|
switch (ime->type) {
|
|
case ip_match_pattern:
|
|
indirect = 0;
|
|
break;
|
|
case ip_match_indirect:
|
|
indirect = 1;
|
|
break;
|
|
case ip_match_localhost:
|
|
ime->u.indirect.list = local_addresses;
|
|
indirect = 1;
|
|
break;
|
|
case ip_match_localnets:
|
|
ime->u.indirect.list = local_networks;
|
|
indirect = 1;
|
|
break;
|
|
case ip_match_key:
|
|
if (key == NULL) {
|
|
indirect = 0;
|
|
break;
|
|
}
|
|
else {
|
|
if (ns_samename(ime->u.key.key->dk_key_name,
|
|
key->dk_key_name) == 1)
|
|
return (1);
|
|
else
|
|
continue;
|
|
}
|
|
default:
|
|
panic("unexpected ime type in ip_match_addr_or_key()",
|
|
NULL);
|
|
}
|
|
if (indirect) {
|
|
ret = ip_match_addr_or_key(ime->u.indirect.list,
|
|
address, key);
|
|
if (ret > 0) {
|
|
if (ime->flags & IP_MATCH_NEGATE)
|
|
ret = (ret) ? 0 : 1;
|
|
return (ret);
|
|
}
|
|
} else {
|
|
if (ina_onnet(address, ime->u.direct.address,
|
|
ime->u.direct.mask)) {
|
|
if (ime->flags & IP_MATCH_NEGATE)
|
|
return (0);
|
|
else
|
|
return (1);
|
|
}
|
|
}
|
|
}
|
|
return (-1);
|
|
}
|
|
|
|
int
|
|
ip_match_address(ip_match_list iml, struct in_addr address) {
|
|
return ip_match_addr_or_key(iml, address, NULL);
|
|
}
|
|
|
|
int
|
|
ip_addr_or_key_allowed(ip_match_list iml, struct in_addr address,
|
|
DST_KEY *key)
|
|
{
|
|
int ret;
|
|
|
|
if (iml == NULL)
|
|
return (0);
|
|
ret = ip_match_addr_or_key(iml, address, key);
|
|
if (ret < 0)
|
|
ret = 0;
|
|
return (ret);
|
|
}
|
|
|
|
int
|
|
ip_address_allowed(ip_match_list iml, struct in_addr address) {
|
|
return(ip_addr_or_key_allowed(iml, address, NULL));
|
|
}
|
|
|
|
int
|
|
ip_match_network(ip_match_list iml, struct in_addr address,
|
|
struct in_addr mask) {
|
|
ip_match_element ime;
|
|
int ret;
|
|
int indirect;
|
|
|
|
INSIST(iml != NULL);
|
|
for (ime = iml->first; ime != NULL; ime = ime->next) {
|
|
switch (ime->type) {
|
|
case ip_match_pattern:
|
|
indirect = 0;
|
|
break;
|
|
case ip_match_indirect:
|
|
indirect = 1;
|
|
break;
|
|
case ip_match_localhost:
|
|
ime->u.indirect.list = local_addresses;
|
|
indirect = 1;
|
|
break;
|
|
case ip_match_localnets:
|
|
ime->u.indirect.list = local_networks;
|
|
indirect = 1;
|
|
break;
|
|
case ip_match_key:
|
|
indirect = 0;
|
|
break;
|
|
default:
|
|
indirect = 0; /* Make gcc happy. */
|
|
panic("unexpected ime type in ip_match_network()",
|
|
NULL);
|
|
}
|
|
if (indirect) {
|
|
ret = ip_match_network(ime->u.indirect.list,
|
|
address, mask);
|
|
if (ret >= 0) {
|
|
if (ime->flags & IP_MATCH_NEGATE)
|
|
ret = (ret) ? 0 : 1;
|
|
return (ret);
|
|
}
|
|
} else {
|
|
if (address.s_addr == ime->u.direct.address.s_addr &&
|
|
mask.s_addr == ime->u.direct.mask.s_addr) {
|
|
if (ime->flags & IP_MATCH_NEGATE)
|
|
return (0);
|
|
else
|
|
return (1);
|
|
}
|
|
}
|
|
}
|
|
return (-1);
|
|
}
|
|
|
|
int
|
|
distance_of_address(ip_match_list iml, struct in_addr address) {
|
|
ip_match_element ime;
|
|
int ret;
|
|
int indirect;
|
|
int distance;
|
|
|
|
INSIST(iml != NULL);
|
|
for (distance = 1, ime = iml->first;
|
|
ime != NULL; ime = ime->next, distance++) {
|
|
switch (ime->type) {
|
|
case ip_match_pattern:
|
|
indirect = 0;
|
|
break;
|
|
case ip_match_indirect:
|
|
indirect = 1;
|
|
break;
|
|
case ip_match_localhost:
|
|
ime->u.indirect.list = local_addresses;
|
|
indirect = 1;
|
|
break;
|
|
case ip_match_localnets:
|
|
ime->u.indirect.list = local_networks;
|
|
indirect = 1;
|
|
break;
|
|
case ip_match_key:
|
|
indirect = 0;
|
|
return (-1);
|
|
default:
|
|
indirect = 0; /* Make gcc happy. */
|
|
panic("unexpected ime type in distance_of_address()",
|
|
NULL);
|
|
}
|
|
if (indirect) {
|
|
ret = ip_match_address(ime->u.indirect.list, address);
|
|
if (ret >= 0) {
|
|
if (ime->flags & IP_MATCH_NEGATE)
|
|
ret = (ret) ? 0 : 1;
|
|
if (distance > MAX_TOPOLOGY_DISTANCE)
|
|
distance = MAX_TOPOLOGY_DISTANCE;
|
|
if (ret)
|
|
return (distance);
|
|
else
|
|
return (MAX_TOPOLOGY_DISTANCE);
|
|
}
|
|
} else {
|
|
if (ina_onnet(address, ime->u.direct.address,
|
|
ime->u.direct.mask)) {
|
|
if (distance > MAX_TOPOLOGY_DISTANCE)
|
|
distance = MAX_TOPOLOGY_DISTANCE;
|
|
if (ime->flags & IP_MATCH_NEGATE)
|
|
return (MAX_TOPOLOGY_DISTANCE);
|
|
else
|
|
return (distance);
|
|
}
|
|
}
|
|
}
|
|
return (UNKNOWN_TOPOLOGY_DISTANCE);
|
|
}
|
|
|
|
int
|
|
ip_match_is_none(ip_match_list iml) {
|
|
ip_match_element ime;
|
|
|
|
if ((iml == NULL) || (iml->first == NULL))
|
|
return (1);
|
|
ime = iml->first;
|
|
if (ime->type == ip_match_indirect) {
|
|
if (ime->flags & IP_MATCH_NEGATE)
|
|
return (0);
|
|
iml = ime->u.indirect.list;
|
|
if ((iml == NULL) || (iml->first == NULL))
|
|
return (0);
|
|
ime = iml->first;
|
|
}
|
|
if (ime->type == ip_match_pattern) {
|
|
if ((ime->flags & IP_MATCH_NEGATE) &&
|
|
ime->u.direct.address.s_addr == 0 &&
|
|
ime->u.direct.mask.s_addr == 0)
|
|
return (1);
|
|
}
|
|
return (0);
|
|
}
|
|
|
|
/*
|
|
* find_forwarder finds the fwddata structure for an address,
|
|
* allocating one if we can't find one already existing.
|
|
*/
|
|
|
|
static struct fwddata *
|
|
find_forwarder(struct in_addr address)
|
|
{
|
|
struct fwddata *fdp;
|
|
struct fwddata **fdpp = NULL;
|
|
struct databuf *ns, *nsdata;
|
|
register int i;
|
|
|
|
for (i = 0; i < fwddata_count; i++) {
|
|
fdp = fwddata[i];
|
|
if (fdp == NULL) {
|
|
if (fdpp == NULL)
|
|
fdpp = &fwddata[i];
|
|
continue;
|
|
}
|
|
if (memcmp(&fdp->fwdaddr.sin_addr, &address,
|
|
sizeof(address)) == 0) {
|
|
fdp->ref_count++;
|
|
return (fdp);
|
|
}
|
|
}
|
|
|
|
fdp = (struct fwddata *)memget(sizeof(struct fwddata));
|
|
if (!fdp)
|
|
panic("memget failed in find_forwarder", NULL);
|
|
|
|
fdp->fwdaddr.sin_family = AF_INET;
|
|
fdp->fwdaddr.sin_addr = address;
|
|
fdp->fwdaddr.sin_port = ns_port;
|
|
|
|
ns = fdp->ns = (struct databuf *)memget(sizeof(*ns));
|
|
if (!ns)
|
|
panic("memget failed in find_forwarder", NULL);
|
|
memset(ns, 0, sizeof(*ns));
|
|
|
|
nsdata = fdp->nsdata = (struct databuf *)memget(sizeof(*nsdata));
|
|
if (!nsdata)
|
|
panic("memget failed in find_forwarder", NULL);
|
|
memset(nsdata, 0, sizeof(*nsdata));
|
|
|
|
ns->d_type = T_NS;
|
|
ns->d_class = C_IN;
|
|
ns->d_rcnt = 1;
|
|
|
|
nsdata->d_type = T_A;
|
|
nsdata->d_class = C_IN;
|
|
nsdata->d_nstime = 1 + (int)(25.0*rand()/(RAND_MAX + 1.0));
|
|
nsdata->d_rcnt = 1;
|
|
|
|
fdp->ref_count = 1;
|
|
|
|
if (fdpp != NULL) {
|
|
*fdpp = fdp;
|
|
return (fdp);
|
|
}
|
|
|
|
i = 0;
|
|
if (fwddata == NULL) {
|
|
fwddata = memget(sizeof *fwddata);
|
|
if (fwddata == NULL)
|
|
i = 1;
|
|
} else {
|
|
register size_t size;
|
|
register struct fwddata **an_tmp;
|
|
|
|
size = fwddata_count * sizeof *fwddata;
|
|
an_tmp = memget(size + sizeof *fwddata);
|
|
if (an_tmp == NULL) {
|
|
i = 1;
|
|
} else {
|
|
memcpy(an_tmp, fwddata, size);
|
|
memput(fwddata, size);
|
|
fwddata = an_tmp;
|
|
}
|
|
}
|
|
|
|
if (i == 0) {
|
|
fwddata[fwddata_count] = fdp;
|
|
fwddata_count++;
|
|
} else {
|
|
ns_warning(ns_log_config, "forwarder add failed (memget) [%s]",
|
|
inet_ntoa(address));
|
|
}
|
|
|
|
return (fdp);
|
|
}
|
|
|
|
/*
|
|
* Forwarder glue
|
|
*
|
|
* XXX This will go away when the rest of bind understands
|
|
* forward zones.
|
|
*/
|
|
|
|
static void
|
|
add_forwarder(struct fwdinfo **fipp, struct in_addr address) {
|
|
struct fwdinfo *fip = *fipp, *ftp = NULL;
|
|
struct fwddata *fdp;
|
|
|
|
#ifdef FWD_LOOP
|
|
if (aIsUs(address)) {
|
|
ns_error(ns_log_config, "forwarder '%s' ignored, my address",
|
|
inet_ntoa(address));
|
|
return;
|
|
}
|
|
#endif /* FWD_LOOP */
|
|
|
|
/* On multiple forwarder lines, move to end of the list. */
|
|
while (fip != NULL && fip->next != NULL)
|
|
fip = fip->next;
|
|
|
|
fdp = find_forwarder(address);
|
|
ftp = (struct fwdinfo *)memget(sizeof(struct fwdinfo));
|
|
if (!ftp)
|
|
panic("memget failed in add_forwarder", NULL);
|
|
ftp->fwddata = fdp;
|
|
ftp->next = NULL;
|
|
if (fip == NULL)
|
|
*fipp = ftp; /* First time only */
|
|
else
|
|
fip->next = ftp;
|
|
}
|
|
|
|
void
|
|
free_also_notify(options op) {
|
|
#ifdef BIND_NOTIFY
|
|
memput(op->also_notify, op->notify_count * sizeof *op->also_notify);
|
|
op->also_notify = NULL;
|
|
op->notify_count = 0;
|
|
#endif
|
|
}
|
|
|
|
int
|
|
add_global_also_notify(options op, struct in_addr address) {
|
|
#ifdef BIND_NOTIFY
|
|
int i;
|
|
|
|
INSIST(op != NULL);
|
|
|
|
ns_debug(ns_log_config, 2, "adding global notify %s",
|
|
inet_ntoa(address));
|
|
|
|
/* Check for duplicates. */
|
|
|
|
for (i = 0; i < op->notify_count; i++) {
|
|
if (memcmp(op->also_notify + i,
|
|
&address, sizeof address) == 0) {
|
|
ns_warning(ns_log_config,
|
|
"duplicate global also-notify address ignored [%s]",
|
|
inet_ntoa(address));
|
|
return (1);
|
|
}
|
|
}
|
|
i = 0;
|
|
|
|
if (op->also_notify == NULL) {
|
|
op->also_notify = memget(sizeof *op->also_notify);
|
|
if (op->also_notify == NULL)
|
|
i = 1;
|
|
} else {
|
|
register size_t size;
|
|
register struct in_addr *an_tmp;
|
|
size = op->notify_count * sizeof *op->also_notify;
|
|
an_tmp = memget(size + sizeof *op->also_notify);
|
|
if (an_tmp == NULL) {
|
|
i = 1;
|
|
} else {
|
|
memcpy(an_tmp, op->also_notify, size);
|
|
memput(op->also_notify, size);
|
|
op->also_notify = an_tmp;
|
|
}
|
|
}
|
|
if (i == 0) {
|
|
op->also_notify[op->notify_count] = address;
|
|
op->notify_count++;
|
|
} else {
|
|
ns_warning(ns_log_config,
|
|
"global also-notify add failed (memget) [%s]",
|
|
inet_ntoa(address));
|
|
}
|
|
#endif
|
|
return (1);
|
|
}
|
|
|
|
void
|
|
add_global_forwarder(options op, struct in_addr address) {
|
|
|
|
INSIST(op != NULL);
|
|
|
|
ns_debug(ns_log_config, 2, "adding default forwarder %s",
|
|
inet_ntoa(address));
|
|
|
|
add_forwarder(&op->fwdtab, address);
|
|
}
|
|
|
|
void
|
|
set_zone_forward(zone_config zh) {
|
|
struct zoneinfo *zp;
|
|
zp = zh.opaque;
|
|
|
|
zp->z_flags |= Z_FORWARD_SET;
|
|
set_zone_boolean_option(zh, OPTION_FORWARD_ONLY, 0);
|
|
}
|
|
|
|
void
|
|
add_zone_forwarder(zone_config zh, struct in_addr address) {
|
|
struct zoneinfo *zp;
|
|
char *zname;
|
|
|
|
zp = zh.opaque;
|
|
INSIST(zp != NULL);
|
|
|
|
zname = (zp->z_origin[0] == '\0') ? "." : zp->z_origin;
|
|
ns_debug(ns_log_config, 2, "adding forwarder %s for zone zone '%s'",
|
|
inet_ntoa(address), zname);
|
|
|
|
zp->z_flags |= Z_FORWARD_SET;
|
|
|
|
add_forwarder(&zp->z_fwdtab, address);
|
|
}
|
|
|
|
void
|
|
free_forwarders(struct fwdinfo *fwdtab) {
|
|
struct fwdinfo *ftp, *fnext;
|
|
int i;
|
|
|
|
for (ftp = fwdtab; ftp != NULL; ftp = fnext) {
|
|
fnext = ftp->next;
|
|
if (--ftp->fwddata->ref_count == 0) {
|
|
for (i = 0 ; i < fwddata_count; i++)
|
|
if (fwddata[i] == ftp->fwddata) {
|
|
fwddata[i] = NULL;
|
|
break;
|
|
}
|
|
memput(ftp->fwddata->ns, sizeof *ftp->fwddata->ns);
|
|
memput(ftp->fwddata->nsdata,
|
|
sizeof *ftp->fwddata->nsdata);
|
|
memput(ftp->fwddata, sizeof *ftp->fwddata);
|
|
}
|
|
memput(ftp, sizeof *ftp);
|
|
}
|
|
fwdtab = NULL;
|
|
}
|
|
|
|
/*
|
|
* Servers
|
|
*/
|
|
|
|
static server_info
|
|
new_server(struct in_addr address) {
|
|
server_info si;
|
|
|
|
si = (server_info)memget(sizeof (struct server_info));
|
|
if (si == NULL)
|
|
panic("memget failed in new_server()", NULL);
|
|
si->address = address;
|
|
si->flags = 0U;
|
|
si->transfers = 0;
|
|
si->transfer_format = axfr_use_default;
|
|
si->key_list = NULL;
|
|
si->next = NULL;
|
|
if (server_options->flags & OPTION_MAINTAIN_IXFR_BASE)
|
|
si->flags |= SERVER_INFO_SUPPORT_IXFR;
|
|
else
|
|
si->flags &= ~SERVER_INFO_SUPPORT_IXFR;
|
|
return (si);
|
|
}
|
|
|
|
static void
|
|
free_server(server_info si) {
|
|
/* Don't free key; it'll be done when the auth table is freed. */
|
|
memput(si, sizeof *si);
|
|
}
|
|
|
|
server_info
|
|
find_server(struct in_addr address) {
|
|
server_info si;
|
|
|
|
for (si = nameserver_info; si != NULL; si = si->next)
|
|
if (si->address.s_addr == address.s_addr)
|
|
break;
|
|
return (si);
|
|
}
|
|
|
|
static void
|
|
add_server(server_info si) {
|
|
ip_match_element ime;
|
|
|
|
si->next = nameserver_info;
|
|
nameserver_info = si;
|
|
|
|
/*
|
|
* To ease transition, we'll add bogus nameservers to an
|
|
* ip matching list. This will probably be redone when the
|
|
* merging of nameserver data structures occurs.
|
|
*/
|
|
if (si->flags & SERVER_INFO_BOGUS) {
|
|
ime = new_ip_match_pattern(si->address, 32);
|
|
INSIST(ime != NULL);
|
|
add_to_ip_match_list(bogus_nameservers, ime);
|
|
}
|
|
ns_debug(ns_log_config, 3, "server %s: flags %08x transfers %d",
|
|
inet_ntoa(si->address), si->flags, si->transfers);
|
|
if (si->key_list != NULL)
|
|
dprint_key_info_list(si->key_list);
|
|
}
|
|
|
|
static void
|
|
free_nameserver_info() {
|
|
server_info si_next, si;
|
|
|
|
for (si = nameserver_info; si != NULL; si = si_next) {
|
|
si_next = si->next;
|
|
free_server(si);
|
|
}
|
|
nameserver_info = NULL;
|
|
if (bogus_nameservers != NULL) {
|
|
free_ip_match_list(bogus_nameservers);
|
|
bogus_nameservers = NULL;
|
|
}
|
|
}
|
|
|
|
static void
|
|
free_secretkey_info() {
|
|
if (secretkey_info != NULL) {
|
|
free_key_info_list(secretkey_info);
|
|
secretkey_info = NULL;
|
|
}
|
|
}
|
|
|
|
server_config
|
|
begin_server(struct in_addr address) {
|
|
server_config sc;
|
|
|
|
sc.opaque = new_server(address);
|
|
return (sc);
|
|
}
|
|
|
|
void
|
|
end_server(server_config sc, int should_install) {
|
|
server_info si;
|
|
|
|
si = sc.opaque;
|
|
|
|
INSIST(si != NULL);
|
|
|
|
if (should_install)
|
|
add_server(si);
|
|
else
|
|
free_server(si);
|
|
sc.opaque = NULL;
|
|
}
|
|
|
|
void
|
|
set_server_option(server_config sc, int bool_opt, int value) {
|
|
server_info si;
|
|
|
|
si = sc.opaque;
|
|
|
|
INSIST(si != NULL);
|
|
|
|
switch (bool_opt) {
|
|
case SERVER_INFO_BOGUS:
|
|
case SERVER_INFO_SUPPORT_IXFR:
|
|
if (value)
|
|
si->flags |= bool_opt;
|
|
else
|
|
si->flags &= ~bool_opt;
|
|
break;
|
|
default:
|
|
panic("unexpected option in set_server_option", NULL);
|
|
}
|
|
}
|
|
|
|
void
|
|
set_server_transfers(server_config sc, int transfers) {
|
|
server_info si;
|
|
|
|
si = sc.opaque;
|
|
|
|
INSIST(si != NULL);
|
|
|
|
if (transfers < 0)
|
|
transfers = 0;
|
|
si->transfers = transfers;
|
|
}
|
|
|
|
void
|
|
set_server_transfer_format(server_config sc,
|
|
enum axfr_format transfer_format) {
|
|
server_info si;
|
|
|
|
si = sc.opaque;
|
|
|
|
INSIST(si != NULL);
|
|
|
|
si->transfer_format = transfer_format;
|
|
}
|
|
|
|
void
|
|
add_server_key_info(server_config sc, DST_KEY *dst_key) {
|
|
server_info si;
|
|
|
|
si = sc.opaque;
|
|
|
|
INSIST(si != NULL);
|
|
|
|
if (si->key_list == NULL)
|
|
si->key_list = new_key_info_list();
|
|
add_to_key_info_list(si->key_list, dst_key);
|
|
}
|
|
|
|
/*
|
|
* Keys
|
|
*/
|
|
|
|
DST_KEY *
|
|
new_key_info(char *name, char *algorithm, char *secret) {
|
|
DST_KEY *dst_key;
|
|
int alg, blen;
|
|
u_char buffer[1024];
|
|
|
|
INSIST(name != NULL);
|
|
INSIST(algorithm != NULL);
|
|
INSIST(secret != NULL);
|
|
alg = tsig_alg_value(algorithm);
|
|
if (alg == -1) {
|
|
ns_warning(ns_log_config, "Unsupported TSIG algorithm %s",
|
|
algorithm);
|
|
return (NULL);
|
|
}
|
|
|
|
blen = b64_pton(secret, buffer, sizeof(buffer));
|
|
if (blen < 0) {
|
|
ns_warning(ns_log_config, "Invalid TSIG secret \"%s\"", secret);
|
|
return (NULL);
|
|
}
|
|
dst_key = dst_buffer_to_key(name, alg,
|
|
NS_KEY_TYPE_AUTH_ONLY|NS_KEY_NAME_ENTITY,
|
|
NS_KEY_PROT_ANY, buffer, blen);
|
|
if (dst_key == NULL)
|
|
ns_warning(ns_log_config,
|
|
"dst_buffer_to_key failed in new_key_info");
|
|
return (dst_key);
|
|
}
|
|
|
|
void
|
|
free_key_info(DST_KEY *dst_key) {
|
|
INSIST(dst_key != NULL);
|
|
dst_free_key(dst_key);
|
|
}
|
|
|
|
DST_KEY *
|
|
find_key(char *name, char *algorithm) {
|
|
key_list_element ke;
|
|
|
|
if (secretkey_info == NULL)
|
|
return (NULL);
|
|
|
|
for (ke = secretkey_info->first; ke != NULL; ke = ke->next) {
|
|
DST_KEY *dst_key = ke->key;
|
|
|
|
if (ns_samename(name, dst_key->dk_key_name) != 1)
|
|
continue;
|
|
if (algorithm == NULL ||
|
|
dst_key->dk_alg == tsig_alg_value(algorithm))
|
|
break;
|
|
}
|
|
if (ke == NULL)
|
|
return (NULL);
|
|
return (ke->key);
|
|
}
|
|
|
|
void
|
|
dprint_key_info(DST_KEY *dst_key) {
|
|
INSIST(dst_key != NULL);
|
|
ns_debug(ns_log_config, 7, "key %s", dst_key->dk_key_name);
|
|
ns_debug(ns_log_config, 7, " algorithm %d", dst_key->dk_alg);
|
|
}
|
|
|
|
key_info_list
|
|
new_key_info_list() {
|
|
key_info_list kil;
|
|
|
|
kil = (key_info_list)memget(sizeof (struct key_info_list));
|
|
if (kil == NULL)
|
|
panic("memget failed in new_key_info_list()", NULL);
|
|
kil->first = NULL;
|
|
kil->last = NULL;
|
|
return (kil);
|
|
}
|
|
|
|
void
|
|
free_key_info_list(key_info_list kil) {
|
|
key_list_element kle, kle_next;
|
|
|
|
INSIST(kil != NULL);
|
|
for (kle = kil->first; kle != NULL; kle = kle_next) {
|
|
kle_next = kle->next;
|
|
/* note we do NOT free kle->info */
|
|
memput(kle, sizeof *kle);
|
|
}
|
|
memput(kil, sizeof *kil);
|
|
}
|
|
|
|
void
|
|
add_to_key_info_list(key_info_list kil, DST_KEY *dst_key) {
|
|
key_list_element kle;
|
|
|
|
INSIST(kil != NULL);
|
|
INSIST(dst_key != NULL);
|
|
|
|
kle = (key_list_element)memget(sizeof (struct key_list_element));
|
|
if (kle == NULL)
|
|
panic("memget failed in add_to_key_info_list()", NULL);
|
|
kle->key = dst_key;
|
|
if (kil->last != NULL)
|
|
kil->last->next = kle;
|
|
kle->next = NULL;
|
|
kil->last = kle;
|
|
if (kil->first == NULL)
|
|
kil->first = kle;
|
|
}
|
|
|
|
void
|
|
dprint_key_info_list(key_info_list kil) {
|
|
key_list_element kle;
|
|
|
|
INSIST(kil != NULL);
|
|
|
|
for (kle = kil->first; kle != NULL; kle = kle->next)
|
|
dprint_key_info(kle->key);
|
|
}
|
|
|
|
/*
|
|
* Logging.
|
|
*/
|
|
|
|
log_config
|
|
begin_logging() {
|
|
log_config log_cfg;
|
|
log_context lc;
|
|
|
|
log_cfg = (log_config)memget(sizeof (struct log_config));
|
|
if (log_cfg == NULL)
|
|
ns_panic(ns_log_config, 0,
|
|
"memget failed creating log_config");
|
|
if (log_new_context(ns_log_max_category, logging_categories, &lc) < 0)
|
|
ns_panic(ns_log_config, 0,
|
|
"log_new_context() failed: %s", strerror(errno));
|
|
log_cfg->log_ctx = lc;
|
|
log_cfg->eventlib_channel = NULL;
|
|
log_cfg->packet_channel = NULL;
|
|
log_cfg->default_debug_active = 0;
|
|
return (log_cfg);
|
|
}
|
|
|
|
void
|
|
add_log_channel(log_config log_cfg, int category, log_channel chan) {
|
|
log_channel_type type;
|
|
|
|
INSIST(log_cfg != NULL);
|
|
|
|
type = log_get_channel_type(chan);
|
|
if (category == ns_log_eventlib) {
|
|
if (type != log_file && type != log_null) {
|
|
ns_error(ns_log_config,
|
|
"must specify a file or null channel for the eventlib category");
|
|
return;
|
|
}
|
|
if (log_cfg->eventlib_channel != NULL) {
|
|
ns_error(ns_log_config,
|
|
"only one channel allowed for the eventlib category");
|
|
return;
|
|
}
|
|
log_cfg->eventlib_channel = chan;
|
|
}
|
|
if (category == ns_log_packet) {
|
|
if (type != log_file && type != log_null) {
|
|
ns_error(ns_log_config,
|
|
"must specify a file or null channel for the packet category");
|
|
return;
|
|
}
|
|
if (log_cfg->packet_channel != NULL) {
|
|
ns_error(ns_log_config,
|
|
"only one channel allowed for the packet category");
|
|
return;
|
|
}
|
|
log_cfg->packet_channel = chan;
|
|
}
|
|
|
|
if (log_add_channel(log_cfg->log_ctx, category, chan) < 0) {
|
|
ns_error(ns_log_config, "log_add_channel() failed");
|
|
return;
|
|
}
|
|
|
|
if (chan == debug_channel)
|
|
log_cfg->default_debug_active = 1;
|
|
}
|
|
|
|
void
|
|
open_special_channels() {
|
|
int using_null = 0;
|
|
|
|
if (log_open_stream(eventlib_channel) == NULL) {
|
|
eventlib_channel = null_channel;
|
|
using_null = 1;
|
|
}
|
|
if (log_open_stream(packet_channel) == NULL) {
|
|
packet_channel = null_channel;
|
|
using_null = 1;
|
|
}
|
|
|
|
if (using_null &&
|
|
log_open_stream(null_channel) == NULL)
|
|
ns_panic(ns_log_config, 1, "couldn't open null channel");
|
|
}
|
|
|
|
void
|
|
set_logging(log_config log_cfg, int is_default) {
|
|
log_context lc;
|
|
|
|
INSIST(log_cfg != NULL);
|
|
lc = log_cfg->log_ctx;
|
|
|
|
/*
|
|
* Add the default category if it's not in the context already.
|
|
*/
|
|
if (!log_category_is_active(lc, ns_log_default)) {
|
|
add_log_channel(log_cfg, ns_log_default, debug_channel);
|
|
add_log_channel(log_cfg, ns_log_default, syslog_channel);
|
|
}
|
|
|
|
/*
|
|
* Add the panic category if it's not in the context already.
|
|
*/
|
|
if (!log_category_is_active(lc, ns_log_panic)) {
|
|
add_log_channel(log_cfg, ns_log_panic, stderr_channel);
|
|
add_log_channel(log_cfg, ns_log_panic, syslog_channel);
|
|
}
|
|
|
|
/*
|
|
* Add the eventlib category if it's not in the context already.
|
|
*/
|
|
if (!log_category_is_active(lc, ns_log_eventlib))
|
|
add_log_channel(log_cfg, ns_log_eventlib, debug_channel);
|
|
|
|
/*
|
|
* Add the packet category if it's not in the context already.
|
|
*/
|
|
if (!log_category_is_active(lc, ns_log_packet))
|
|
add_log_channel(log_cfg, ns_log_packet, debug_channel);
|
|
|
|
#ifdef DEBUG
|
|
/*
|
|
* Preserve debugging state.
|
|
*/
|
|
log_option(lc, LOG_OPTION_DEBUG, debug);
|
|
log_option(lc, LOG_OPTION_LEVEL, debug);
|
|
#endif
|
|
|
|
/*
|
|
* Special case for query-log, so we can co-exist with the command
|
|
* line option and SIGWINCH.
|
|
*/
|
|
if (log_category_is_active(lc, ns_log_queries))
|
|
qrylog = 1;
|
|
|
|
/*
|
|
* Cleanup the old context.
|
|
*/
|
|
if (need_logging_free)
|
|
log_free_context(log_ctx);
|
|
|
|
/*
|
|
* The default file channels will never have their reference counts
|
|
* drop to zero, and so they will not be closed by the logging system
|
|
* when log_free_context() is called. We don't want to keep files
|
|
* open unnecessarily, and we want them to behave like user-created
|
|
* channels, so we close them here.
|
|
*/
|
|
if (log_get_stream(debug_channel) != stderr)
|
|
(void)log_close_stream(debug_channel);
|
|
(void)log_close_stream(null_channel);
|
|
|
|
/*
|
|
* Install the new context.
|
|
*/
|
|
log_ctx = lc;
|
|
eventlib_channel = log_cfg->eventlib_channel;
|
|
packet_channel = log_cfg->packet_channel;
|
|
|
|
#ifdef DEBUG
|
|
if (debug) {
|
|
open_special_channels();
|
|
evSetDebug(ev, debug, log_get_stream(eventlib_channel));
|
|
}
|
|
#endif
|
|
|
|
log_ctx_valid = 1;
|
|
need_logging_free = 1;
|
|
logging_installed = 1;
|
|
default_logging_installed = is_default;
|
|
}
|
|
|
|
void
|
|
end_logging(log_config log_cfg, int should_install) {
|
|
if (should_install)
|
|
set_logging(log_cfg, 0);
|
|
else
|
|
log_free_context(log_cfg->log_ctx);
|
|
memput(log_cfg, sizeof (struct log_config));
|
|
}
|
|
|
|
void
|
|
use_default_logging() {
|
|
log_config log_cfg;
|
|
|
|
log_cfg = begin_logging();
|
|
set_logging(log_cfg, 1);
|
|
memput(log_cfg, sizeof (struct log_config));
|
|
}
|
|
|
|
static void
|
|
init_default_log_channels() {
|
|
u_int flags;
|
|
char *name;
|
|
FILE *stream;
|
|
|
|
syslog_channel = log_new_syslog_channel(0, log_info, ISC_FACILITY);
|
|
if (syslog_channel == NULL || log_inc_references(syslog_channel) < 0)
|
|
ns_panic(ns_log_config, 0, "couldn't create syslog_channel");
|
|
|
|
flags = LOG_USE_CONTEXT_LEVEL|LOG_REQUIRE_DEBUG;
|
|
if (foreground) {
|
|
name = NULL;
|
|
stream = stderr;
|
|
} else {
|
|
name = _PATH_DEBUG;
|
|
stream = NULL;
|
|
}
|
|
debug_channel = log_new_file_channel(flags, log_info, name, stream,
|
|
0, ULONG_MAX);
|
|
if (debug_channel == NULL || log_inc_references(debug_channel) < 0)
|
|
ns_panic(ns_log_config, 0, "couldn't create debug_channel");
|
|
log_set_file_owner(debug_channel, user_id, group_id);
|
|
|
|
stderr_channel = log_new_file_channel(0, log_info, NULL, stderr,
|
|
0, ULONG_MAX);
|
|
if (stderr_channel == NULL || log_inc_references(stderr_channel) < 0)
|
|
ns_panic(ns_log_config, 0, "couldn't create stderr_channel");
|
|
log_set_file_owner(stderr_channel, user_id, group_id);
|
|
|
|
null_channel = log_new_file_channel(LOG_CHANNEL_OFF, log_info,
|
|
_PATH_DEVNULL, NULL, 0, ULONG_MAX);
|
|
if (null_channel == NULL || log_inc_references(null_channel) < 0)
|
|
ns_panic(ns_log_config, 0, "couldn't create null_channel");
|
|
log_set_file_owner(null_channel, user_id, group_id);
|
|
}
|
|
|
|
static void
|
|
shutdown_default_log_channels() {
|
|
log_free_channel(syslog_channel);
|
|
log_free_channel(debug_channel);
|
|
log_free_channel(stderr_channel);
|
|
log_free_channel(null_channel);
|
|
}
|
|
|
|
void
|
|
init_logging() {
|
|
int size;
|
|
const struct ns_sym *s;
|
|
char category_name[256];
|
|
|
|
size = ns_log_max_category * (sizeof (char *));
|
|
|
|
logging_categories = (char **)memget(size);
|
|
if (logging_categories == NULL)
|
|
ns_panic(ns_log_config, 0, "memget failed in init_logging");
|
|
memset(logging_categories, 0, size);
|
|
for (s = category_constants; s != NULL && s->name != NULL; s++) {
|
|
sprintf(category_name, "%s: ", s->name);
|
|
logging_categories[s->number] = savestr(category_name, 1);
|
|
}
|
|
|
|
init_default_log_channels();
|
|
use_default_logging();
|
|
}
|
|
|
|
void
|
|
shutdown_logging() {
|
|
int size;
|
|
const struct ns_sym *s;
|
|
|
|
evSetDebug(ev, 0, NULL);
|
|
shutdown_default_log_channels();
|
|
log_free_context(log_ctx);
|
|
|
|
for (s = category_constants; s != NULL && s->name != NULL; s++)
|
|
freestr(logging_categories[s->number]);
|
|
size = ns_log_max_category * (sizeof (char *));
|
|
memput(logging_categories, size);
|
|
logging_categories = NULL;
|
|
}
|
|
|
|
/*
|
|
* Main Loader
|
|
*/
|
|
|
|
void
|
|
init_configuration() {
|
|
/*
|
|
* Remember initial limits for use if "default" is specified in
|
|
* a config file.
|
|
*/
|
|
#ifdef HAVE_GETRUSAGE
|
|
get_initial_limits();
|
|
#endif
|
|
zone_symbol_table = new_symbol_table(ZONE_SYM_TABLE_SIZE, NULL);
|
|
use_default_options();
|
|
parser_initialize();
|
|
ns_ctl_initialize();
|
|
config_initialized = 1;
|
|
}
|
|
|
|
void
|
|
shutdown_configuration() {
|
|
REQUIRE(config_initialized);
|
|
|
|
ns_ctl_shutdown();
|
|
if (server_options != NULL) {
|
|
free_options(server_options);
|
|
server_options = NULL;
|
|
}
|
|
if (current_pid_filename != NULL)
|
|
freestr(current_pid_filename);
|
|
free_nameserver_info();
|
|
free_secretkey_info();
|
|
free_symbol_table(zone_symbol_table);
|
|
parser_shutdown();
|
|
config_initialized = 0;
|
|
}
|
|
|
|
time_t
|
|
load_configuration(const char *filename) {
|
|
time_t mtime;
|
|
|
|
REQUIRE(config_initialized);
|
|
|
|
ns_debug(ns_log_config, 3, "load configuration %s", filename);
|
|
|
|
loading = 1;
|
|
|
|
/*
|
|
* Clean up any previous configuration and initialize
|
|
* global data structures we'll be updating.
|
|
*/
|
|
free_nameserver_info();
|
|
free_secretkey_info();
|
|
bogus_nameservers = new_ip_match_list();
|
|
|
|
options_installed = 0;
|
|
logging_installed = 0;
|
|
|
|
mtime = parse_configuration(filename);
|
|
|
|
/*
|
|
* If the user didn't specify logging or options, but they previously
|
|
* had specified one or both of them, then we need to
|
|
* re-establish the default environment. We have to be careful
|
|
* about when we install default options because the parser
|
|
* must respect limits (e.g. data-size, number of open files)
|
|
* specified in the options file. In the ordinary case where the
|
|
* options section isn't changing on a zone reload, it would be bad
|
|
* to lower these limits temporarily, because we might not survive
|
|
* to the point where they get raised back again. The logging case
|
|
* has similar motivation -- we don't want to override the existing
|
|
* logging scheme (perhaps causing log messages to go somewhere
|
|
* unexpected) when the user hasn't expressed a desire for a new
|
|
* scheme.
|
|
*/
|
|
if (!logging_installed)
|
|
use_default_logging();
|
|
if (!options_installed && !default_options_installed) {
|
|
use_default_options();
|
|
ns_warning(ns_log_config, "re-establishing default options");
|
|
}
|
|
|
|
update_pid_file();
|
|
|
|
/* Init or reinit the interface/port list and associated sockets. */
|
|
getnetconf(0);
|
|
opensocket_f();
|
|
|
|
initial_configuration = 0;
|
|
loading = 0;
|
|
/* release queued notifies */
|
|
notify_afterload();
|
|
return (mtime);
|
|
}
|