summaryrefslogtreecommitdiffstats
path: root/lib/dns/zone.c
diff options
context:
space:
mode:
Diffstat (limited to 'lib/dns/zone.c')
-rw-r--r--lib/dns/zone.c3742
1 files changed, 3607 insertions, 135 deletions
diff --git a/lib/dns/zone.c b/lib/dns/zone.c
index 36f303c..423b005 100644
--- a/lib/dns/zone.c
+++ b/lib/dns/zone.c
@@ -1,5 +1,5 @@
/*
- * Copyright (C) 2004-2008 Internet Systems Consortium, Inc. ("ISC")
+ * Copyright (C) 2004-2009 Internet Systems Consortium, Inc. ("ISC")
* Copyright (C) 1999-2003 Internet Software Consortium.
*
* Permission to use, copy, modify, and/or distribute this software for any
@@ -15,11 +15,12 @@
* PERFORMANCE OF THIS SOFTWARE.
*/
-/* $Id: zone.c,v 1.410.18.55 2008/10/24 01:43:17 tbox Exp $ */
+/* $Id: zone.c,v 1.483.36.6 2009/03/26 22:57:07 marka Exp $ */
/*! \file */
#include <config.h>
+#include <errno.h>
#include <isc/file.h>
#include <isc/mutex.h>
@@ -29,6 +30,9 @@
#include <isc/refcount.h>
#include <isc/rwlock.h>
#include <isc/serial.h>
+#include <isc/strerror.h>
+#include <isc/stats.h>
+#include <isc/stdtime.h>
#include <isc/string.h>
#include <isc/taskpool.h>
#include <isc/timer.h>
@@ -40,29 +44,37 @@
#include <dns/callbacks.h>
#include <dns/db.h>
#include <dns/dbiterator.h>
+#include <dns/dnssec.h>
#include <dns/events.h>
#include <dns/journal.h>
+#include <dns/keyvalues.h>
#include <dns/log.h>
#include <dns/master.h>
#include <dns/masterdump.h>
#include <dns/message.h>
#include <dns/name.h>
+#include <dns/nsec.h>
+#include <dns/nsec3.h>
#include <dns/peer.h>
#include <dns/rcode.h>
#include <dns/rdataclass.h>
#include <dns/rdatalist.h>
#include <dns/rdataset.h>
+#include <dns/rdatasetiter.h>
#include <dns/rdatastruct.h>
#include <dns/rdatatype.h>
#include <dns/request.h>
#include <dns/resolver.h>
#include <dns/result.h>
-#include <dns/stats.h>
+#include <dns/soa.h>
#include <dns/ssu.h>
+#include <dns/stats.h>
#include <dns/tsig.h>
#include <dns/xfrin.h>
#include <dns/zone.h>
+#include <dst/dst.h>
+
#define ZONE_MAGIC ISC_MAGIC('Z', 'O', 'N', 'E')
#define DNS_ZONE_VALID(zone) ISC_MAGIC_VALID(zone, ZONE_MAGIC)
@@ -90,6 +102,8 @@
#define RANGE(a, min, max) \
(((a) < (min)) ? (min) : ((a) < (max) ? (a) : (max)))
+#define NSEC3REMOVE(x) (((x) & DNS_NSEC3FLAG_REMOVE) != 0)
+
/*
* Default values.
*/
@@ -111,6 +125,10 @@ typedef struct dns_load dns_load_t;
typedef struct dns_forward dns_forward_t;
typedef struct dns_io dns_io_t;
typedef ISC_LIST(dns_io_t) dns_iolist_t;
+typedef struct dns_signing dns_signing_t;
+typedef ISC_LIST(dns_signing_t) dns_signinglist_t;
+typedef struct dns_nsec3chain dns_nsec3chain_t;
+typedef ISC_LIST(dns_nsec3chain_t) dns_nsec3chainlist_t;
#define DNS_ZONE_CHECKLOCK
#ifdef DNS_ZONE_CHECKLOCK
@@ -178,11 +196,16 @@ struct dns_zone {
isc_time_t dumptime;
isc_time_t loadtime;
isc_time_t notifytime;
+ isc_time_t resigntime;
+ isc_time_t keywarntime;
+ isc_time_t signingtime;
+ isc_time_t nsec3chaintime;
isc_uint32_t serial;
isc_uint32_t refresh;
isc_uint32_t retry;
isc_uint32_t expire;
isc_uint32_t minimum;
+ isc_stdtime_t key_expiry;
char *keydirectory;
isc_uint32_t maxrefresh;
@@ -215,6 +238,7 @@ struct dns_zone {
dns_acl_t *forward_acl;
dns_acl_t *notify_acl;
dns_acl_t *query_acl;
+ dns_acl_t *queryon_acl;
dns_acl_t *xfr_acl;
isc_boolean_t update_disabled;
isc_boolean_t zero_no_soa_ttl;
@@ -232,6 +256,7 @@ struct dns_zone {
isc_event_t ctlevent;
dns_ssutable_t *ssutable;
isc_uint32_t sigvalidityinterval;
+ isc_uint32_t sigresigninginterval;
dns_view_t *view;
dns_acache_t *acache;
dns_checkmxfunc_t checkmx;
@@ -247,17 +272,39 @@ struct dns_zone {
ISC_LINK(dns_zone_t) statelink;
dns_zonelist_t *statelist;
/*%
- * Optional per-zone statistics counters (NULL if not present).
+ * Statistics counters about zone management.
*/
- isc_uint64_t *counters;
+ isc_stats_t *stats;
+ /*%
+ * Optional per-zone statistics counters. Counted outside of this
+ * module.
+ */
+ isc_boolean_t requeststats_on;
+ isc_stats_t *requeststats;
isc_uint32_t notifydelay;
dns_isselffunc_t isself;
void *isselfarg;
+ char * strnamerd;
+ char * strname;
+ char * strrdclass;
+ char * strviewname;
+
/*%
* Serial number for deferred journal compaction.
*/
isc_uint32_t compact_serial;
+ /*%
+ * Keys that are signing the zone for the first time.
+ */
+ dns_signinglist_t signing;
+ dns_nsec3chainlist_t nsec3chain;
+ /*%
+ * Signing / re-signing quantum stopping parameters.
+ */
+ isc_uint32_t signatures;
+ isc_uint32_t nodes;
+ dns_rdatatype_t privatetype;
};
#define DNS_ZONE_FLAG(z,f) (ISC_TF(((z)->flags & (f)) != 0))
@@ -287,7 +334,7 @@ struct dns_zone {
* reload */
#define DNS_ZONEFLG_NOMASTERS 0x00001000U /*%< an attempt to refresh a
* zone with no masters
- * occured */
+ * occurred */
#define DNS_ZONEFLG_LOADING 0x00002000U /*%< load from disk in progress*/
#define DNS_ZONEFLG_HAVETIMERS 0x00004000U /*%< timer values have been set
* from SOA (if not set, we
@@ -310,6 +357,21 @@ struct dns_zone {
/* Flags for zone_load() */
#define DNS_ZONELOADFLAG_NOSTAT 0x00000001U /* Do not stat() master files */
+#define UNREACH_CHACHE_SIZE 10U
+#define UNREACH_HOLD_TIME 600 /* 10 minutes */
+
+#define CHECK(op) \
+ do { result = (op); \
+ if (result != ISC_R_SUCCESS) goto failure; \
+ } while (0)
+
+struct dns_unreachable {
+ isc_sockaddr_t remote;
+ isc_sockaddr_t local;
+ isc_uint32_t expire;
+ isc_uint32_t last;
+};
+
struct dns_zonemgr {
unsigned int magic;
isc_mem_t * mctx;
@@ -338,6 +400,10 @@ struct dns_zonemgr {
isc_uint32_t ioactive;
dns_iolist_t high;
dns_iolist_t low;
+
+ /* Locked by rwlock. */
+ /* LRU cache */
+ struct dns_unreachable unreachable[UNREACH_CHACHE_SIZE];
};
/*%
@@ -410,6 +476,56 @@ struct dns_io {
isc_event_t *event;
};
+/*%
+ * Hold state for when we are signing a zone with a new
+ * DNSKEY as result of an update.
+ */
+struct dns_signing {
+ unsigned int magic;
+ dns_db_t *db;
+ dns_dbiterator_t *dbiterator;
+ dns_secalg_t algorithm;
+ isc_uint16_t keyid;
+ isc_boolean_t delete;
+ isc_boolean_t done;
+ ISC_LINK(dns_signing_t) link;
+};
+
+struct dns_nsec3chain {
+ unsigned int magic;
+ dns_db_t *db;
+ dns_dbiterator_t *dbiterator;
+ dns_rdata_nsec3param_t nsec3param;
+ unsigned char salt[255];
+ isc_boolean_t done;
+ isc_boolean_t seen_nsec;
+ isc_boolean_t delete_nsec;
+ isc_boolean_t save_delete_nsec;
+ ISC_LINK(dns_nsec3chain_t) link;
+};
+/*%<
+ * 'dbiterator' contains a iterator for the database. If we are creating
+ * a NSEC3 chain only the non-NSEC3 nodes will be iterated. If we are
+ * removing a NSEC3 chain then both NSEC3 and non-NSEC3 nodes will be
+ * iterated.
+ *
+ * 'nsec3param' contains the parameters of the NSEC3 chain being created
+ * or removed.
+ *
+ * 'salt' is buffer space and is referenced via 'nsec3param.salt'.
+ *
+ * 'seen_nsec' will be set to true if, while iterating the zone to create a
+ * NSEC3 chain, a NSEC record is seen.
+ *
+ * 'delete_nsec' will be set to true if, at the completion of the creation
+ * of a NSEC3 chain, 'seen_nsec' is true. If 'delete_nsec' is true then we
+ * are in the process of deleting the NSEC chain.
+ *
+ * 'save_delete_nsec' is used to store the initial state of 'delete_nsec'
+ * so it can be recovered in the event of a error.
+ */
+
+
#define SEND_BUFFER_SIZE 2048
static void zone_settimer(dns_zone_t *, isc_time_t *);
@@ -436,6 +552,10 @@ static void zone_shutdown(isc_task_t *, isc_event_t *);
static void zone_loaddone(void *arg, isc_result_t result);
static isc_result_t zone_startload(dns_db_t *db, dns_zone_t *zone,
isc_time_t loadtime);
+static void zone_namerd_tostr(dns_zone_t *zone, char *buf, size_t length);
+static void zone_name_tostr(dns_zone_t *zone, char *buf, size_t length);
+static void zone_rdclass_tostr(dns_zone_t *zone, char *buf, size_t length);
+static void zone_viewname_tostr(dns_zone_t *zone, char *buf, size_t length);
#if 0
/* ondestroy example */
@@ -484,6 +604,12 @@ static void zone_saveunique(dns_zone_t *zone, const char *path,
static void zone_maintenance(dns_zone_t *zone);
static void zone_notify(dns_zone_t *zone, isc_time_t *now);
static void dump_done(void *arg, isc_result_t result);
+static isc_boolean_t dns_zonemgr_unreachable(dns_zonemgr_t *zmgr,
+ isc_sockaddr_t *remote,
+ isc_sockaddr_t *local,
+ isc_time_t *now);
+static isc_result_t zone_signwithkey(dns_zone_t *zone, dns_secalg_t algorithm,
+ isc_uint16_t keyid, isc_boolean_t delete);
#define ENTER zone_debuglog(zone, me, 1, "enter")
@@ -518,6 +644,15 @@ static const char *dbargv_default[] = { "rbt" };
} \
} while (0)
+/*%
+ * Increment resolver-related statistics counters. Zone must be locked.
+ */
+static inline void
+inc_stats(dns_zone_t *zone, isc_statscounter_t counter) {
+ if (zone->stats != NULL)
+ isc_stats_increment(zone->stats, counter);
+}
+
/***
*** Public functions.
***/
@@ -559,8 +694,12 @@ dns_zone_create(dns_zone_t **zonep, isc_mem_t *mctx) {
goto free_dblock;
zone->irefs = 0;
dns_name_init(&zone->origin, NULL);
+ zone->strnamerd = NULL;
+ zone->strname = NULL;
+ zone->strrdclass = NULL;
+ zone->strviewname = NULL;
zone->masterfile = NULL;
- zone->masterformat = dns_masterformat_none;
+ zone->masterformat = dns_masterformat_none;
zone->keydirectory = NULL;
zone->journalsize = -1;
zone->journal = NULL;
@@ -575,6 +714,10 @@ dns_zone_create(dns_zone_t **zonep, isc_mem_t *mctx) {
isc_time_settoepoch(&zone->dumptime);
isc_time_settoepoch(&zone->loadtime);
zone->notifytime = now;
+ isc_time_settoepoch(&zone->resigntime);
+ isc_time_settoepoch(&zone->keywarntime);
+ isc_time_settoepoch(&zone->signingtime);
+ isc_time_settoepoch(&zone->nsec3chaintime);
zone->serial = 0;
zone->refresh = DNS_ZONE_DEFAULTREFRESH;
zone->retry = DNS_ZONE_DEFAULTRETRY;
@@ -597,6 +740,7 @@ dns_zone_create(dns_zone_t **zonep, isc_mem_t *mctx) {
zone->forward_acl = NULL;
zone->notify_acl = NULL;
zone->query_acl = NULL;
+ zone->queryon_acl = NULL;
zone->xfr_acl = NULL;
zone->update_disabled = ISC_FALSE;
zone->zero_no_soa_ttl = ISC_TRUE;
@@ -622,6 +766,7 @@ dns_zone_create(dns_zone_t **zonep, isc_mem_t *mctx) {
zone->maxxfrout = MAX_XFER_TIME;
zone->ssutable = NULL;
zone->sigvalidityinterval = 30 * 24 * 3600;
+ zone->sigresigninginterval = 7 * 24 * 3600;
zone->view = NULL;
zone->acache = NULL;
zone->checkmx = NULL;
@@ -629,10 +774,17 @@ dns_zone_create(dns_zone_t **zonep, isc_mem_t *mctx) {
zone->checkns = NULL;
ISC_LINK_INIT(zone, statelink);
zone->statelist = NULL;
- zone->counters = NULL;
+ zone->stats = NULL;
+ zone->requeststats_on = ISC_FALSE;
+ zone->requeststats = NULL;
zone->notifydelay = 5;
zone->isself = NULL;
zone->isselfarg = NULL;
+ ISC_LIST_INIT(zone->signing);
+ ISC_LIST_INIT(zone->nsec3chain);
+ zone->signatures = 10;
+ zone->nodes = 100;
+ zone->privatetype = (dns_rdatatype_t)0xffffU;
zone->magic = ZONE_MAGIC;
@@ -669,6 +821,8 @@ dns_zone_create(dns_zone_t **zonep, isc_mem_t *mctx) {
static void
zone_free(dns_zone_t *zone) {
isc_mem_t *mctx = NULL;
+ dns_signing_t *signing;
+ dns_nsec3chain_t *nsec3chain;
REQUIRE(DNS_ZONE_VALID(zone));
REQUIRE(isc_refcount_current(&zone->erefs) == 0);
@@ -687,10 +841,26 @@ zone_free(dns_zone_t *zone) {
if (zone->task != NULL)
isc_task_detach(&zone->task);
- if (zone->zmgr)
+ if (zone->zmgr != NULL)
dns_zonemgr_releasezone(zone->zmgr, zone);
/* Unmanaged objects */
+ for (signing = ISC_LIST_HEAD(zone->signing);
+ signing != NULL;
+ signing = ISC_LIST_HEAD(zone->signing)) {
+ ISC_LIST_UNLINK(zone->signing, signing, link);
+ dns_db_detach(&signing->db);
+ dns_dbiterator_destroy(&signing->dbiterator);
+ isc_mem_put(zone->mctx, signing, sizeof *signing);
+ }
+ for (nsec3chain = ISC_LIST_HEAD(zone->nsec3chain);
+ nsec3chain != NULL;
+ nsec3chain = ISC_LIST_HEAD(zone->nsec3chain)) {
+ ISC_LIST_UNLINK(zone->nsec3chain, nsec3chain, link);
+ dns_db_detach(&nsec3chain->db);
+ dns_dbiterator_destroy(&nsec3chain->dbiterator);
+ isc_mem_put(zone->mctx, nsec3chain, sizeof *nsec3chain);
+ }
if (zone->masterfile != NULL)
isc_mem_free(zone->mctx, zone->masterfile);
zone->masterfile = NULL;
@@ -701,8 +871,10 @@ zone_free(dns_zone_t *zone) {
if (zone->journal != NULL)
isc_mem_free(zone->mctx, zone->journal);
zone->journal = NULL;
- if (zone->counters != NULL)
- dns_stats_freecounters(zone->mctx, &zone->counters);
+ if (zone->stats != NULL)
+ isc_stats_detach(&zone->stats);
+ if (zone->requeststats != NULL)
+ isc_stats_detach(&zone->requeststats);
if (zone->db != NULL)
zone_detachdb(zone);
if (zone->acache != NULL)
@@ -721,10 +893,20 @@ zone_free(dns_zone_t *zone) {
dns_acl_detach(&zone->notify_acl);
if (zone->query_acl != NULL)
dns_acl_detach(&zone->query_acl);
+ if (zone->queryon_acl != NULL)
+ dns_acl_detach(&zone->queryon_acl);
if (zone->xfr_acl != NULL)
dns_acl_detach(&zone->xfr_acl);
if (dns_name_dynamic(&zone->origin))
dns_name_free(&zone->origin, zone->mctx);
+ if (zone->strnamerd != NULL)
+ isc_mem_free(zone->mctx, zone->strnamerd);
+ if (zone->strname != NULL)
+ isc_mem_free(zone->mctx, zone->strname);
+ if (zone->strrdclass != NULL)
+ isc_mem_free(zone->mctx, zone->strrdclass);
+ if (zone->strviewname != NULL)
+ isc_mem_free(zone->mctx, zone->strviewname);
if (zone->ssutable != NULL)
dns_ssutable_detach(&zone->ssutable);
@@ -743,6 +925,7 @@ zone_free(dns_zone_t *zone) {
*/
void
dns_zone_setclass(dns_zone_t *zone, dns_rdataclass_t rdclass) {
+ char namebuf[1024];
REQUIRE(DNS_ZONE_VALID(zone));
REQUIRE(rdclass != dns_rdataclass_none);
@@ -754,11 +937,22 @@ dns_zone_setclass(dns_zone_t *zone, dns_rdataclass_t rdclass) {
REQUIRE(zone->rdclass == dns_rdataclass_none ||
zone->rdclass == rdclass);
zone->rdclass = rdclass;
+
+ if (zone->strnamerd != NULL)
+ isc_mem_free(zone->mctx, zone->strnamerd);
+ if (zone->strrdclass != NULL)
+ isc_mem_free(zone->mctx, zone->strrdclass);
+
+ zone_namerd_tostr(zone, namebuf, sizeof namebuf);
+ zone->strnamerd = isc_mem_strdup(zone->mctx, namebuf);
+ zone_rdclass_tostr(zone, namebuf, sizeof namebuf);
+ zone->strrdclass = isc_mem_strdup(zone->mctx, namebuf);
+
UNLOCK_ZONE(zone);
}
dns_rdataclass_t
-dns_zone_getclass(dns_zone_t *zone){
+dns_zone_getclass(dns_zone_t *zone) {
REQUIRE(DNS_ZONE_VALID(zone));
return (zone->rdclass);
@@ -773,6 +967,19 @@ dns_zone_setnotifytype(dns_zone_t *zone, dns_notifytype_t notifytype) {
UNLOCK_ZONE(zone);
}
+isc_uint32_t
+dns_zone_getserial(dns_zone_t *zone) {
+ isc_uint32_t serial;
+
+ REQUIRE(DNS_ZONE_VALID(zone));
+
+ LOCK_ZONE(zone);
+ serial = zone->serial;
+ UNLOCK_ZONE(zone);
+
+ return (serial);
+}
+
/*
* Single shot.
*/
@@ -888,12 +1095,24 @@ dns_zone_setdbtype(dns_zone_t *zone,
void
dns_zone_setview(dns_zone_t *zone, dns_view_t *view) {
+ char namebuf[1024];
REQUIRE(DNS_ZONE_VALID(zone));
LOCK_ZONE(zone);
if (zone->view != NULL)
dns_view_weakdetach(&zone->view);
dns_view_weakattach(view, &zone->view);
+
+ if (zone->strviewname != NULL)
+ isc_mem_free(zone->mctx, zone->strviewname);
+ if (zone->strnamerd != NULL)
+ isc_mem_free(zone->mctx, zone->strnamerd);
+
+ zone_namerd_tostr(zone, namebuf, sizeof namebuf);
+ zone->strnamerd = isc_mem_strdup(zone->mctx, namebuf);
+ zone_viewname_tostr(zone, namebuf, sizeof namebuf);
+ zone->strviewname = isc_mem_strdup(zone->mctx, namebuf);
+
UNLOCK_ZONE(zone);
}
@@ -909,6 +1128,7 @@ dns_zone_getview(dns_zone_t *zone) {
isc_result_t
dns_zone_setorigin(dns_zone_t *zone, const dns_name_t *origin) {
isc_result_t result;
+ char namebuf[1024];
REQUIRE(DNS_ZONE_VALID(zone));
REQUIRE(origin != NULL);
@@ -919,6 +1139,17 @@ dns_zone_setorigin(dns_zone_t *zone, const dns_name_t *origin) {
dns_name_init(&zone->origin, NULL);
}
result = dns_name_dup(origin, zone->mctx, &zone->origin);
+
+ if (zone->strnamerd != NULL)
+ isc_mem_free(zone->mctx, zone->strnamerd);
+ if (zone->strname != NULL)
+ isc_mem_free(zone->mctx, zone->strname);
+
+ zone_namerd_tostr(zone, namebuf, sizeof namebuf);
+ zone->strnamerd = isc_mem_strdup(zone->mctx, namebuf);
+ zone_name_tostr(zone, namebuf, sizeof namebuf);
+ zone->strname = isc_mem_strdup(zone->mctx, namebuf);
+
UNLOCK_ZONE(zone);
return (result);
}
@@ -1064,11 +1295,7 @@ zone_isdynamic(dns_zone_t *zone) {
zone->type == dns_zone_stub ||
(!zone->update_disabled && zone->ssutable != NULL) ||
(!zone->update_disabled && zone->update_acl != NULL &&
- ! (zone->update_acl->length == 1 &&
- zone->update_acl->elements[0].negative == ISC_TRUE
- &&
- zone->update_acl->elements[0].type ==
- dns_aclelementtype_any))));
+ !dns_acl_isnone(zone->update_acl))));
}
@@ -1243,6 +1470,33 @@ dns_zone_loadnew(dns_zone_t *zone) {
return (zone_load(zone, DNS_ZONELOADFLAG_NOSTAT));
}
+static unsigned int
+get_master_options(dns_zone_t *zone) {
+ unsigned int options;
+
+ options = DNS_MASTER_ZONE;
+ if (zone->type == dns_zone_slave)
+ options |= DNS_MASTER_SLAVE;
+ if (DNS_ZONE_OPTION(zone, DNS_ZONEOPT_CHECKNS))
+ options |= DNS_MASTER_CHECKNS;
+ if (DNS_ZONE_OPTION(zone, DNS_ZONEOPT_FATALNS))
+ options |= DNS_MASTER_FATALNS;
+ if (DNS_ZONE_OPTION(zone, DNS_ZONEOPT_CHECKNAMES))
+ options |= DNS_MASTER_CHECKNAMES;
+ if (DNS_ZONE_OPTION(zone, DNS_ZONEOPT_CHECKNAMESFAIL))
+ options |= DNS_MASTER_CHECKNAMESFAIL;
+ if (DNS_ZONE_OPTION(zone, DNS_ZONEOPT_CHECKMX))
+ options |= DNS_MASTER_CHECKMX;
+ if (DNS_ZONE_OPTION(zone, DNS_ZONEOPT_CHECKMXFAIL))
+ options |= DNS_MASTER_CHECKMXFAIL;
+ if (DNS_ZONE_OPTION(zone, DNS_ZONEOPT_CHECKWILDCARD))
+ options |= DNS_MASTER_CHECKWILDCARD;
+ if (zone->type == dns_zone_master &&
+ (zone->update_acl != NULL || zone->ssutable != NULL))
+ options |= DNS_MASTER_RESIGN;
+ return (options);
+}
+
static void
zone_gotreadhandle(isc_task_t *task, isc_event_t *event) {
dns_load_t *load = event->ev_arg;
@@ -1257,28 +1511,14 @@ zone_gotreadhandle(isc_task_t *task, isc_event_t *event) {
if (result == ISC_R_CANCELED)
goto fail;
- options = DNS_MASTER_ZONE;
- if (load->zone->type == dns_zone_slave)
- options |= DNS_MASTER_SLAVE;
- if (DNS_ZONE_OPTION(load->zone, DNS_ZONEOPT_CHECKNS))
- options |= DNS_MASTER_CHECKNS;
- if (DNS_ZONE_OPTION(load->zone, DNS_ZONEOPT_FATALNS))
- options |= DNS_MASTER_FATALNS;
- if (DNS_ZONE_OPTION(load->zone, DNS_ZONEOPT_CHECKNAMES))
- options |= DNS_MASTER_CHECKNAMES;
- if (DNS_ZONE_OPTION(load->zone, DNS_ZONEOPT_CHECKNAMESFAIL))
- options |= DNS_MASTER_CHECKNAMESFAIL;
- if (DNS_ZONE_OPTION(load->zone, DNS_ZONEOPT_CHECKMX))
- options |= DNS_MASTER_CHECKMX;
- if (DNS_ZONE_OPTION(load->zone, DNS_ZONEOPT_CHECKMXFAIL))
- options |= DNS_MASTER_CHECKMXFAIL;
- if (DNS_ZONE_OPTION(load->zone, DNS_ZONEOPT_CHECKWILDCARD))
- options |= DNS_MASTER_CHECKWILDCARD;
- result = dns_master_loadfileinc2(load->zone->masterfile,
+ options = get_master_options(load->zone);
+
+ result = dns_master_loadfileinc3(load->zone->masterfile,
dns_db_origin(load->db),
dns_db_origin(load->db),
load->zone->rdclass,
options,
+ load->zone->sigresigninginterval,
&load->callbacks, task,
zone_loaddone, load,
&load->zone->lctx, load->zone->mctx,
@@ -1334,25 +1574,10 @@ zone_startload(dns_db_t *db, dns_zone_t *zone, isc_time_t loadtime) {
isc_result_t tresult;
unsigned int options;
- options = DNS_MASTER_ZONE;
+ options = get_master_options(zone);
+
if (DNS_ZONE_OPTION(zone, DNS_ZONEOPT_MANYERRORS))
options |= DNS_MASTER_MANYERRORS;
- if (DNS_ZONE_OPTION(zone, DNS_ZONEOPT_CHECKNS))
- options |= DNS_MASTER_CHECKNS;
- if (DNS_ZONE_OPTION(zone, DNS_ZONEOPT_FATALNS))
- options |= DNS_MASTER_FATALNS;
- if (zone->type == dns_zone_slave)
- options |= DNS_MASTER_SLAVE;
- if (DNS_ZONE_OPTION(zone, DNS_ZONEOPT_CHECKNAMES))
- options |= DNS_MASTER_CHECKNAMES;
- if (DNS_ZONE_OPTION(zone, DNS_ZONEOPT_CHECKNAMESFAIL))
- options |= DNS_MASTER_CHECKNAMESFAIL;
- if (DNS_ZONE_OPTION(zone, DNS_ZONEOPT_CHECKMX))
- options |= DNS_MASTER_CHECKMX;
- if (DNS_ZONE_OPTION(zone, DNS_ZONEOPT_CHECKMXFAIL))
- options |= DNS_MASTER_CHECKMXFAIL;
- if (DNS_ZONE_OPTION(zone, DNS_ZONEOPT_CHECKWILDCARD))
- options |= DNS_MASTER_CHECKWILDCARD;
if (zone->zmgr != NULL && zone->db != NULL && zone->task != NULL) {
load = isc_mem_get(zone->mctx, sizeof(*load));
@@ -1394,9 +1619,10 @@ zone_startload(dns_db_t *db, dns_zone_t *zone, isc_time_t loadtime) {
&callbacks.add_private);
if (result != ISC_R_SUCCESS)
return (result);
- result = dns_master_loadfile2(zone->masterfile, &zone->origin,
+ result = dns_master_loadfile3(zone->masterfile, &zone->origin,
&zone->origin, zone->rdclass,
- options, &callbacks, zone->mctx,
+ options, zone->sigresigninginterval,
+ &callbacks, zone->mctx,
zone->masterformat);
tresult = dns_db_endload(db, &callbacks.add_private);
if (result == ISC_R_SUCCESS)
@@ -1727,7 +1953,7 @@ integrity_checks(dns_zone_t *zone, dns_db_t *db) {
dns_rdataset_init(&rdataset);
dns_rdata_init(&rdata);
- result = dns_db_createiterator(db, ISC_FALSE, &dbiterator);
+ result = dns_db_createiterator(db, 0, &dbiterator);
if (result != ISC_R_SUCCESS)
return (ISC_TRUE);
@@ -1890,6 +2116,292 @@ zone_check_dnskeys(dns_zone_t *zone, dns_db_t *db) {
}
+static void
+resume_signingwithkey(dns_zone_t *zone) {
+ dns_dbnode_t *node = NULL;
+ dns_dbversion_t *version = NULL;
+ dns_rdata_t rdata = DNS_RDATA_INIT;
+ dns_rdataset_t rdataset;
+ isc_result_t result;
+
+ result = dns_db_findnode(zone->db, &zone->origin, ISC_FALSE, &node);
+ if (result != ISC_R_SUCCESS)
+ goto cleanup;
+
+ dns_db_currentversion(zone->db, &version);
+ dns_rdataset_init(&rdataset);
+ result = dns_db_findrdataset(zone->db, node, version,
+ zone->privatetype,
+ dns_rdatatype_none, 0,
+ &rdataset, NULL);
+ if (result != ISC_R_SUCCESS)
+ goto cleanup;
+
+ for (result = dns_rdataset_first(&rdataset);
+ result == ISC_R_SUCCESS;
+ result = dns_rdataset_next(&rdataset))
+ {
+ dns_rdataset_current(&rdataset, &rdata);
+ if (rdata.length != 5 || rdata.data[4] != 0) {
+ dns_rdata_reset(&rdata);
+ continue;
+ }
+
+ result = zone_signwithkey(zone, rdata.data[0],
+ (rdata.data[1] << 8) | rdata.data[2], ISC_TF(rdata.data[3]));
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "zone_signwithkey failed: %s",
+ dns_result_totext(result));
+ }
+ dns_rdata_reset(&rdata);
+ }
+ dns_rdataset_disassociate(&rdataset);
+
+ cleanup:
+ if (node != NULL)
+ dns_db_detachnode(zone->db, &node);
+ if (version != NULL)
+ dns_db_closeversion(zone->db, &version, ISC_FALSE);
+
+}
+
+static isc_result_t
+zone_addnsec3chain(dns_zone_t *zone, dns_rdata_nsec3param_t *nsec3param) {
+ dns_nsec3chain_t *nsec3chain, *current;
+ isc_result_t result;
+ isc_time_t now;
+ unsigned int options = 0;
+
+ nsec3chain = isc_mem_get(zone->mctx, sizeof *nsec3chain);
+ if (nsec3chain == NULL)
+ return (ISC_R_NOMEMORY);
+
+ nsec3chain->magic = 0;
+ nsec3chain->done = ISC_FALSE;
+ nsec3chain->db = NULL;
+ nsec3chain->dbiterator = NULL;
+ nsec3chain->nsec3param.common.rdclass = nsec3param->common.rdclass;
+ nsec3chain->nsec3param.common.rdtype = nsec3param->common.rdtype;
+ nsec3chain->nsec3param.hash = nsec3param->hash;
+ nsec3chain->nsec3param.iterations = nsec3param->iterations;
+ nsec3chain->nsec3param.flags = nsec3param->flags;
+ nsec3chain->nsec3param.salt_length = nsec3param->salt_length;
+ memcpy(nsec3chain->salt, nsec3param->salt, nsec3param->salt_length);
+ nsec3chain->nsec3param.salt = nsec3chain->salt;
+ nsec3chain->seen_nsec = ISC_FALSE;
+ nsec3chain->delete_nsec = ISC_FALSE;
+ nsec3chain->save_delete_nsec = ISC_FALSE;
+
+ for (current = ISC_LIST_HEAD(zone->nsec3chain);
+ current != NULL;
+ current = ISC_LIST_NEXT(current, link)) {
+ if (current->db == zone->db &&
+ current->nsec3param.hash == nsec3param->hash &&
+ current->nsec3param.iterations == nsec3param->iterations &&
+ current->nsec3param.salt_length == nsec3param->salt_length
+ && !memcmp(current->nsec3param.salt, nsec3param->salt,
+ nsec3param->salt_length))
+ current->done = ISC_TRUE;
+ }
+
+ if (zone->db != NULL) {
+ dns_db_attach(zone->db, &nsec3chain->db);
+ if ((nsec3chain->nsec3param.flags & DNS_NSEC3FLAG_CREATE) != 0)
+ options = DNS_DB_NONSEC3;
+ result = dns_db_createiterator(nsec3chain->db, options,
+ &nsec3chain->dbiterator);
+ if (result == ISC_R_SUCCESS)
+ dns_dbiterator_first(nsec3chain->dbiterator);
+ if (result == ISC_R_SUCCESS) {
+ dns_dbiterator_pause(nsec3chain->dbiterator);
+ ISC_LIST_INITANDAPPEND(zone->nsec3chain,
+ nsec3chain, link);
+ nsec3chain = NULL;
+ if (isc_time_isepoch(&zone->nsec3chaintime)) {
+ TIME_NOW(&now);
+ zone->nsec3chaintime = now;
+ if (zone->task != NULL)
+ zone_settimer(zone, &now);
+ }
+ }
+ } else
+ result = ISC_R_NOTFOUND;
+
+ if (nsec3chain != NULL) {
+ if (nsec3chain->db != NULL)
+ dns_db_detach(&nsec3chain->db);
+ if (nsec3chain->dbiterator != NULL)
+ dns_dbiterator_destroy(&nsec3chain->dbiterator);
+ isc_mem_put(zone->mctx, nsec3chain, sizeof *nsec3chain);
+ }
+ return (result);
+}
+
+static void
+resume_addnsec3chain(dns_zone_t *zone) {
+ dns_dbnode_t *node = NULL;
+ dns_dbversion_t *version = NULL;
+ dns_rdata_t rdata = DNS_RDATA_INIT;
+ dns_rdataset_t rdataset;
+ isc_result_t result;
+ dns_rdata_nsec3param_t nsec3param;
+
+ result = dns_db_findnode(zone->db, &zone->origin, ISC_FALSE, &node);
+ if (result != ISC_R_SUCCESS)
+ goto cleanup;
+
+ dns_db_currentversion(zone->db, &version);
+ dns_rdataset_init(&rdataset);
+ result = dns_db_findrdataset(zone->db, node, version,
+ dns_rdatatype_nsec3param,
+ dns_rdatatype_none, 0,
+ &rdataset, NULL);
+ if (result != ISC_R_SUCCESS)
+ goto cleanup;
+
+ for (result = dns_rdataset_first(&rdataset);
+ result == ISC_R_SUCCESS;
+ result = dns_rdataset_next(&rdataset))
+ {
+ dns_rdataset_current(&rdataset, &rdata);
+ result = dns_rdata_tostruct(&rdata, &nsec3param, NULL);
+ RUNTIME_CHECK(result == ISC_R_SUCCESS);
+ if ((nsec3param.flags & DNS_NSEC3FLAG_CREATE) != 0 ||
+ (nsec3param.flags & DNS_NSEC3FLAG_REMOVE) != 0) {
+ result = zone_addnsec3chain(zone, &nsec3param);
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "zone_addnsec3chain failed: %s",
+ dns_result_totext(result));
+ }
+ }
+ dns_rdata_reset(&rdata);
+ }
+ dns_rdataset_disassociate(&rdataset);
+
+ cleanup:
+ if (node != NULL)
+ dns_db_detachnode(zone->db, &node);
+ if (version != NULL)
+ dns_db_closeversion(zone->db, &version, ISC_FALSE);
+}
+
+static void
+set_resigntime(dns_zone_t *zone) {
+ dns_rdataset_t rdataset;
+ dns_fixedname_t fixed;
+ char namebuf[DNS_NAME_FORMATSIZE];
+ unsigned int resign;
+ isc_result_t result;
+ isc_uint32_t nanosecs;
+
+ dns_rdataset_init(&rdataset);
+ dns_fixedname_init(&fixed);
+ result = dns_db_getsigningtime(zone->db, &rdataset,
+ dns_fixedname_name(&fixed));
+ if (result != ISC_R_SUCCESS) {
+ isc_time_settoepoch(&zone->resigntime);
+ return;
+ }
+ resign = rdataset.resign;
+ dns_name_format(dns_fixedname_name(&fixed), namebuf, sizeof(namebuf));
+ dns_rdataset_disassociate(&rdataset);
+ isc_random_get(&nanosecs);
+ nanosecs %= 1000000000;
+ isc_time_set(&zone->resigntime, resign, nanosecs);
+}
+
+static isc_result_t
+check_nsec3param(dns_zone_t *zone, dns_db_t *db) {
+ dns_dbnode_t *node = NULL;
+ dns_rdataset_t rdataset;
+ dns_dbversion_t *version = NULL;
+ dns_rdata_nsec3param_t nsec3param;
+ isc_boolean_t ok = ISC_FALSE;
+ isc_result_t result;
+ dns_rdata_t rdata = DNS_RDATA_INIT;
+ isc_boolean_t dynamic = (zone->type == dns_zone_master) ?
+ zone_isdynamic(zone) : ISC_FALSE;
+
+ dns_rdataset_init(&rdataset);
+ result = dns_db_findnode(db, &zone->origin, ISC_FALSE, &node);
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "nsec3param lookup failure: %s",
+ dns_result_totext(result));
+ return (result);
+ }
+ dns_db_currentversion(db, &version);
+
+ result = dns_db_findrdataset(db, node, version,
+ dns_rdatatype_nsec3param,
+ dns_rdatatype_none, 0, &rdataset, NULL);
+ if (result == ISC_R_NOTFOUND) {
+ result = ISC_R_SUCCESS;
+ goto cleanup;
+ }
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "nsec3param lookup failure: %s",
+ dns_result_totext(result));
+ goto cleanup;
+ }
+
+ /*
+ * For dynamic zones we must support every algorithm so we can
+ * regenerate all the NSEC3 chains.
+ * For non-dynamic zones we only need to find a supported algorithm.
+ */
+ for (result = dns_rdataset_first(&rdataset);
+ result == ISC_R_SUCCESS;
+ result = dns_rdataset_next(&rdataset))
+ {
+ dns_rdataset_current(&rdataset, &rdata);
+ result = dns_rdata_tostruct(&rdata, &nsec3param, NULL);
+ dns_rdata_reset(&rdata);
+ INSIST(result == ISC_R_SUCCESS);
+ if (DNS_ZONE_OPTION(zone, DNS_ZONEOPT_NSEC3TESTZONE) &&
+ nsec3param.hash == DNS_NSEC3_UNKNOWNALG && !dynamic)
+ {
+ dns_zone_log(zone, ISC_LOG_WARNING,
+ "nsec3 test \"unknown\" hash algorithm found: %u",
+ nsec3param.hash);
+ ok = ISC_TRUE;
+ } else if (!dns_nsec3_supportedhash(nsec3param.hash)) {
+ if (dynamic) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "unsupported nsec3 hash algorithm"
+ " in dynamic zone: %u",
+ nsec3param.hash);
+ result = DNS_R_BADZONE;
+ /* Stop second error message. */
+ ok = ISC_TRUE;
+ break;
+ } else
+ dns_zone_log(zone, ISC_LOG_WARNING,
+ "unsupported nsec3 hash algorithm: %u",
+ nsec3param.hash);
+ } else
+ ok = ISC_TRUE;
+ }
+ if (result == ISC_R_NOMORE)
+ result = ISC_R_SUCCESS;
+
+ if (!ok) {
+ result = DNS_R_BADZONE;
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "no supported nsec3 hash algorithm");
+ }
+
+ cleanup:
+ if (dns_rdataset_isassociated(&rdataset))
+ dns_rdataset_disassociate(&rdataset);
+ dns_db_closeversion(db, &version, ISC_FALSE);
+ dns_db_detachnode(db, &node);
+ return (result);
+}
+
static isc_result_t
zone_postload(dns_zone_t *zone, dns_db_t *db, isc_time_t loadtime,
isc_result_t result)
@@ -1901,6 +2413,7 @@ zone_postload(dns_zone_t *zone, dns_db_t *db, isc_time_t loadtime,
isc_time_t now;
isc_boolean_t needdump = ISC_FALSE;
isc_boolean_t hasinclude = DNS_ZONE_FLAG(zone, DNS_ZONEFLG_HASINCLUDE);
+ unsigned int options;
TIME_NOW(&now);
@@ -1945,7 +2458,12 @@ zone_postload(dns_zone_t *zone, dns_db_t *db, isc_time_t loadtime,
! DNS_ZONE_OPTION(zone, DNS_ZONEOPT_NOMERGE) &&
! DNS_ZONE_FLAG(zone, DNS_ZONEFLG_LOADED))
{
- result = dns_journal_rollforward(zone->mctx, db,
+ if (zone->type == dns_zone_master &&
+ (zone->update_acl != NULL || zone->ssutable != NULL))
+ options = DNS_JOURNALOPT_RESIGN;
+ else
+ options = 0;
+ result = dns_journal_rollforward(zone->mctx, db, options,
zone->journal);
if (result != ISC_R_SUCCESS && result != ISC_R_NOTFOUND &&
result != DNS_R_UPTODATE && result != DNS_R_NOJOURNAL &&
@@ -1972,7 +2490,6 @@ zone_postload(dns_zone_t *zone, dns_db_t *db, isc_time_t loadtime,
zone->loadtime = loadtime;
dns_zone_log(zone, ISC_LOG_DEBUG(1), "loaded");
-
/*
* Obtain ns, soa and cname counts for top of zone.
*/
@@ -2010,6 +2527,11 @@ zone_postload(dns_zone_t *zone, dns_db_t *db, isc_time_t loadtime,
result = DNS_R_BADZONE;
goto cleanup;
}
+ if (zone->type != dns_zone_stub) {
+ result = check_nsec3param(zone, db);
+ if (result != ISC_R_SUCCESS)
+ goto cleanup;
+ }
if (zone->type == dns_zone_master &&
DNS_ZONE_OPTION(zone, DNS_ZONEOPT_CHECKINTEGRITY) &&
!integrity_checks(zone, db)) {
@@ -2047,6 +2569,17 @@ zone_postload(dns_zone_t *zone, dns_db_t *db, isc_time_t loadtime,
"zone may fail to transfer "
"to slaves.");
}
+
+ if (zone->type == dns_zone_master &&
+ (zone->update_acl != NULL || zone->ssutable != NULL) &&
+ zone->sigresigninginterval < (3 * refresh) &&
+ dns_db_issecure(db))
+ {
+ dns_zone_log(zone, ISC_LOG_WARNING,
+ "sig-re-signing-interval less than "
+ "3 * refresh.");
+ }
+
zone->serial = serial;
zone->refresh = RANGE(refresh,
zone->minrefresh, zone->maxrefresh);
@@ -2121,8 +2654,14 @@ zone_postload(dns_zone_t *zone, dns_db_t *db, isc_time_t loadtime,
result = ISC_R_SUCCESS;
if (needdump)
zone_needdump(zone, DNS_DUMP_DELAY);
- if (zone->task != NULL)
+ if (zone->task != NULL) {
+ if (zone->type == dns_zone_master) {
+ set_resigntime(zone);
+ resume_signingwithkey(zone);
+ resume_addnsec3chain(zone);
+ }
zone_settimer(zone, &now);
+ }
if (! dns_db_ispersistent(db))
dns_zone_log(zone, ISC_LOG_INFO, "loaded serial %u%s",
@@ -2809,7 +3348,7 @@ dns_zone_setmasterswithkeys(dns_zone_t *zone,
goto unlock;
/*
- * masters must countain count elements!
+ * masters must contain count elements!
*/
new = isc_mem_get(zone->mctx, count * sizeof(*new));
if (new == NULL) {
@@ -2935,6 +3474,2432 @@ was_dumping(dns_zone_t *zone) {
return (dumping);
}
+#define MAXZONEKEYS 10
+
+static isc_result_t
+do_one_tuple(dns_difftuple_t **tuple, dns_db_t *db, dns_dbversion_t *ver,
+ dns_diff_t *diff)
+{
+ dns_diff_t temp_diff;
+ isc_result_t result;
+
+ /*
+ * Create a singleton diff.
+ */
+ dns_diff_init(diff->mctx, &temp_diff);
+ temp_diff.resign = diff->resign;
+ ISC_LIST_APPEND(temp_diff.tuples, *tuple, link);
+
+ /*
+ * Apply it to the database.
+ */
+ result = dns_diff_apply(&temp_diff, db, ver);
+ ISC_LIST_UNLINK(temp_diff.tuples, *tuple, link);
+ if (result != ISC_R_SUCCESS) {
+ dns_difftuple_free(tuple);
+ return (result);
+ }
+
+ /*
+ * Merge it into the current pending journal entry.
+ */
+ dns_diff_appendminimal(diff, tuple);
+
+ /*
+ * Do not clear temp_diff.
+ */
+ return (ISC_R_SUCCESS);
+}
+
+static isc_result_t
+increment_soa_serial(dns_db_t *db, dns_dbversion_t *ver,
+ dns_diff_t *diff, isc_mem_t *mctx)
+{
+ dns_difftuple_t *deltuple = NULL;
+ dns_difftuple_t *addtuple = NULL;
+ isc_uint32_t serial;
+ isc_result_t result;
+
+ CHECK(dns_db_createsoatuple(db, ver, mctx, DNS_DIFFOP_DEL, &deltuple));
+ CHECK(dns_difftuple_copy(deltuple, &addtuple));
+ addtuple->op = DNS_DIFFOP_ADD;
+
+ serial = dns_soa_getserial(&addtuple->rdata);
+
+ /* RFC1982 */
+ serial = (serial + 1) & 0xFFFFFFFF;
+ if (serial == 0)
+ serial = 1;
+
+ dns_soa_setserial(serial, &addtuple->rdata);
+ CHECK(do_one_tuple(&deltuple, db, ver, diff));
+ CHECK(do_one_tuple(&addtuple, db, ver, diff));
+ result = ISC_R_SUCCESS;
+
+ failure:
+ if (addtuple != NULL)
+ dns_difftuple_free(&addtuple);
+ if (deltuple != NULL)
+ dns_difftuple_free(&deltuple);
+ return (result);
+}
+
+static isc_result_t
+update_one_rr(dns_db_t *db, dns_dbversion_t *ver, dns_diff_t *diff,
+ dns_diffop_t op, dns_name_t *name, dns_ttl_t ttl,
+ dns_rdata_t *rdata)
+{
+ dns_difftuple_t *tuple = NULL;
+ isc_result_t result;
+ result = dns_difftuple_create(diff->mctx, op,
+ name, ttl, rdata, &tuple);
+ if (result != ISC_R_SUCCESS)
+ return (result);
+ return (do_one_tuple(&tuple, db, ver, diff));
+}
+
+static isc_boolean_t
+ksk_sanity(dns_db_t *db, dns_dbversion_t *ver) {
+ isc_boolean_t ret = ISC_FALSE;
+ isc_boolean_t have_ksk = ISC_FALSE, have_nonksk = ISC_FALSE;
+ isc_result_t result;
+ dns_dbnode_t *node = NULL;
+ dns_rdataset_t rdataset;
+ dns_rdata_t rdata = DNS_RDATA_INIT;
+ dns_rdata_dnskey_t dnskey;
+
+ dns_rdataset_init(&rdataset);
+ CHECK(dns_db_findnode(db, dns_db_origin(db), ISC_FALSE, &node));
+ CHECK(dns_db_findrdataset(db, node, ver, dns_rdatatype_dnskey, 0, 0,
+ &rdataset, NULL));
+ CHECK(dns_rdataset_first(&rdataset));
+ while (result == ISC_R_SUCCESS && (!have_ksk || !have_nonksk)) {
+ dns_rdataset_current(&rdataset, &rdata);
+ CHECK(dns_rdata_tostruct(&rdata, &dnskey, NULL));
+ if ((dnskey.flags & (DNS_KEYFLAG_OWNERMASK|DNS_KEYTYPE_NOAUTH))
+ == DNS_KEYOWNER_ZONE) {
+ if ((dnskey.flags & DNS_KEYFLAG_KSK) != 0)
+ have_ksk = ISC_TRUE;
+ else
+ have_nonksk = ISC_TRUE;
+ }
+ dns_rdata_reset(&rdata);
+ result = dns_rdataset_next(&rdataset);
+ }
+ if (have_ksk && have_nonksk)
+ ret = ISC_TRUE;
+ failure:
+ if (dns_rdataset_isassociated(&rdataset))
+ dns_rdataset_disassociate(&rdataset);
+ if (node != NULL)
+ dns_db_detachnode(db, &node);
+ return (ret);
+}
+
+static isc_result_t
+find_zone_keys(dns_zone_t *zone, dns_db_t *db, dns_dbversion_t *ver,
+ isc_mem_t *mctx, unsigned int maxkeys,
+ dst_key_t **keys, unsigned int *nkeys)
+{
+ isc_result_t result;
+ dns_dbnode_t *node = NULL;
+ const char *directory = dns_zone_getkeydirectory(zone);
+ CHECK(dns_db_findnode(db, dns_db_origin(db), ISC_FALSE, &node));
+ result = dns_dnssec_findzonekeys2(db, ver, node, dns_db_origin(db),
+ directory, mctx, maxkeys, keys,
+ nkeys);
+ if (result == ISC_R_NOTFOUND)
+ result = ISC_R_SUCCESS;
+ failure:
+ if (node != NULL)
+ dns_db_detachnode(db, &node);
+ return (result);
+}
+
+static isc_result_t
+offline(dns_db_t *db, dns_dbversion_t *ver, dns_diff_t *diff, dns_name_t *name,
+ dns_ttl_t ttl, dns_rdata_t *rdata)
+{
+ isc_result_t result;
+
+ if ((rdata->flags & DNS_RDATA_OFFLINE) != 0)
+ return (ISC_R_SUCCESS);
+ result = update_one_rr(db, ver, diff, DNS_DIFFOP_DELRESIGN,
+ name, ttl, rdata);
+ if (result != ISC_R_SUCCESS)
+ return (result);
+ rdata->flags |= DNS_RDATA_OFFLINE;
+ result = update_one_rr(db, ver, diff, DNS_DIFFOP_ADDRESIGN,
+ name, ttl, rdata);
+ return (result);
+}
+
+static void
+set_key_expiry_warning(dns_zone_t *zone, isc_stdtime_t when, isc_stdtime_t now)
+{
+ unsigned int delta;
+
+ zone->key_expiry = when;
+ if (when <= now) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "DNSKEY RRSIG(s) have expired");
+ isc_time_settoepoch(&zone->keywarntime);
+ } else if (when < now + 7 * 24 * 3600) {
+ dns_zone_log(zone, ISC_LOG_WARNING,
+ "DNSKEY RRSIG(s) will expire at %u",
+ when); /* XXXMPA convert to date. */
+ delta = when - now;
+ delta--; /* loop prevention */
+ delta /= 24 * 3600; /* to whole days */
+ delta *= 24 * 3600; /* to seconds */
+ isc_time_set(&zone->keywarntime, when - delta, 0);
+ } else {
+ dns_zone_log(zone, ISC_LOG_NOTICE, /* XXMPA ISC_LOG_DEBUG(1) */
+ "setting keywarntime to %u - 7 days",
+ when); /* XXXMPA convert to date. */
+ isc_time_set(&zone->keywarntime, when - 7 * 24 * 3600, 0);
+ }
+}
+
+/*
+ * Delete expired RRsigs and any RRsigs we are about to re-sign.
+ * See also update.c:del_keysigs().
+ */
+static isc_result_t
+del_sigs(dns_zone_t *zone, dns_db_t *db, dns_dbversion_t *ver, dns_name_t *name,
+ dns_rdatatype_t type, dns_diff_t *diff, dst_key_t **keys,
+ unsigned int nkeys, isc_stdtime_t now)
+{
+ isc_result_t result;
+ dns_dbnode_t *node = NULL;
+ dns_rdataset_t rdataset;
+ dns_rdata_t rdata = DNS_RDATA_INIT;
+ unsigned int i;
+ dns_rdata_rrsig_t rrsig;
+ isc_boolean_t found;
+ isc_stdtime_t warn = 0, maybe = 0;
+
+ dns_rdataset_init(&rdataset);
+
+ if (type == dns_rdatatype_nsec3)
+ result = dns_db_findnsec3node(db, name, ISC_FALSE, &node);
+ else
+ result = dns_db_findnode(db, name, ISC_FALSE, &node);
+ if (result == ISC_R_NOTFOUND)
+ return (ISC_R_SUCCESS);
+ if (result != ISC_R_SUCCESS)
+ goto failure;
+ result = dns_db_findrdataset(db, node, ver, dns_rdatatype_rrsig, type,
+ (isc_stdtime_t) 0, &rdataset, NULL);
+ dns_db_detachnode(db, &node);
+
+ if (result == ISC_R_NOTFOUND)
+ return (ISC_R_SUCCESS);
+ if (result != ISC_R_SUCCESS)
+ goto failure;
+
+ for (result = dns_rdataset_first(&rdataset);
+ result == ISC_R_SUCCESS;
+ result = dns_rdataset_next(&rdataset)) {
+ dns_rdataset_current(&rdataset, &rdata);
+ result = dns_rdata_tostruct(&rdata, &rrsig, NULL);
+ RUNTIME_CHECK(result == ISC_R_SUCCESS);
+
+ if (type != dns_rdatatype_dnskey) {
+ result = update_one_rr(db, ver, diff,
+ DNS_DIFFOP_DEL, name,
+ rdataset.ttl, &rdata);
+ dns_rdata_reset(&rdata);
+ if (result != ISC_R_SUCCESS)
+ break;
+ continue;
+ }
+
+ /*
+ * RRSIG(DNSKEY) requires special processing.
+ */
+ found = ISC_FALSE;
+ for (i = 0; i < nkeys; i++) {
+ if (rrsig.algorithm == dst_key_alg(keys[i]) &&
+ rrsig.keyid == dst_key_id(keys[i])) {
+ found = ISC_TRUE;
+ /*
+ * Mark offline RRSIG(DNSKEY).
+ * We want the earliest offline expire time
+ * iff there is a new offline signature.
+ */
+ if (!dst_key_isprivate(keys[i])) {
+ if (warn != 0 &&
+ warn > rrsig.timeexpire)
+ warn = rrsig.timeexpire;
+ if (rdata.flags & DNS_RDATA_OFFLINE) {
+ if (maybe == 0 ||
+ maybe > rrsig.timeexpire)
+ maybe =
+ rrsig.timeexpire;
+ break;
+ }
+ if (warn == 0)
+ warn = maybe;
+ if (warn == 0 ||
+ warn > rrsig.timeexpire)
+ warn = rrsig.timeexpire;
+ result = offline(db, ver, diff, name,
+ rdataset.ttl, &rdata);
+ break;
+ }
+ result = update_one_rr(db, ver, diff,
+ DNS_DIFFOP_DEL,
+ name, rdataset.ttl,
+ &rdata);
+ break;
+ }
+ }
+ /*
+ * If there is not a matching DNSKEY then
+ * delete the RRSIG.
+ */
+ if (!found)
+ result = update_one_rr(db, ver, diff, DNS_DIFFOP_DEL,
+ name, rdataset.ttl, &rdata);
+ dns_rdata_reset(&rdata);
+ if (result != ISC_R_SUCCESS)
+ break;
+ }
+ dns_rdataset_disassociate(&rdataset);
+ if (result == ISC_R_NOMORE)
+ result = ISC_R_SUCCESS;
+ if (warn != 0)
+ set_key_expiry_warning(zone, warn, now);
+ failure:
+ if (node != NULL)
+ dns_db_detachnode(db, &node);
+ return (result);
+}
+
+static isc_result_t
+add_sigs(dns_db_t *db, dns_dbversion_t *ver, dns_name_t *name,
+ dns_rdatatype_t type, dns_diff_t *diff, dst_key_t **keys,
+ unsigned int nkeys, isc_mem_t *mctx, isc_stdtime_t inception,
+ isc_stdtime_t expire, isc_boolean_t check_ksk)
+{
+ isc_result_t result;
+ dns_dbnode_t *node = NULL;
+ dns_rdataset_t rdataset;
+ dns_rdata_t sig_rdata = DNS_RDATA_INIT;
+ unsigned char data[1024]; /* XXX */
+ isc_buffer_t buffer;
+ unsigned int i;
+
+ dns_rdataset_init(&rdataset);
+ isc_buffer_init(&buffer, data, sizeof(data));
+
+ if (type == dns_rdatatype_nsec3)
+ result = dns_db_findnsec3node(db, name, ISC_FALSE, &node);
+ else
+ result = dns_db_findnode(db, name, ISC_FALSE, &node);
+ if (result == ISC_R_NOTFOUND)
+ return (ISC_R_SUCCESS);
+ if (result != ISC_R_SUCCESS)
+ goto failure;
+ result = dns_db_findrdataset(db, node, ver, type, 0,
+ (isc_stdtime_t) 0, &rdataset, NULL);
+ dns_db_detachnode(db, &node);
+ if (result == ISC_R_NOTFOUND)
+ return (ISC_R_SUCCESS);
+ if (result != ISC_R_SUCCESS)
+ goto failure;
+
+ for (i = 0; i < nkeys; i++) {
+ if (check_ksk && type != dns_rdatatype_dnskey &&
+ (dst_key_flags(keys[i]) & DNS_KEYFLAG_KSK) != 0)
+ continue;
+ if (!dst_key_isprivate(keys[i]))
+ continue;
+ /* Calculate the signature, creating a RRSIG RDATA. */
+ CHECK(dns_dnssec_sign(name, &rdataset, keys[i],
+ &inception, &expire,
+ mctx, &buffer, &sig_rdata));
+ /* Update the database and journal with the RRSIG. */
+ /* XXX inefficient - will cause dataset merging */
+ CHECK(update_one_rr(db, ver, diff, DNS_DIFFOP_ADDRESIGN,
+ name, rdataset.ttl, &sig_rdata));
+ dns_rdata_reset(&sig_rdata);
+ }
+
+ failure:
+ if (dns_rdataset_isassociated(&rdataset))
+ dns_rdataset_disassociate(&rdataset);
+ if (node != NULL)
+ dns_db_detachnode(db, &node);
+ return (result);
+}
+
+static void
+zone_resigninc(dns_zone_t *zone) {
+ const char *journalfile;
+ dns_db_t *db = NULL;
+ dns_dbversion_t *version = NULL;
+ dns_diff_t sig_diff;
+ dns_fixedname_t fixed;
+ dns_name_t *name;
+ dns_rdataset_t rdataset;
+ dns_rdatatype_t covers;
+ dst_key_t *zone_keys[MAXZONEKEYS];
+ isc_boolean_t check_ksk;
+ isc_result_t result;
+ isc_stdtime_t now, inception, soaexpire, expire, stop;
+ isc_uint32_t jitter;
+ unsigned int i;
+ unsigned int nkeys = 0;
+ unsigned int resign;
+
+ dns_rdataset_init(&rdataset);
+ dns_fixedname_init(&fixed);
+ dns_diff_init(zone->mctx, &sig_diff);
+ sig_diff.resign = zone->sigresigninginterval;
+
+ /*
+ * Updates are disabled. Pause for 5 minutes.
+ */
+ if (zone->update_disabled) {
+ result = ISC_R_FAILURE;
+ goto failure;
+ }
+
+ ZONEDB_LOCK(&zone->dblock, isc_rwlocktype_read);
+ dns_db_attach(zone->db, &db);
+ ZONEDB_UNLOCK(&zone->dblock, isc_rwlocktype_read);
+
+ result = dns_db_newversion(db, &version);
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "zone_resigninc:dns_db_newversion -> %s\n",
+ dns_result_totext(result));
+ goto failure;
+ }
+
+ result = find_zone_keys(zone, db, version, zone->mctx, MAXZONEKEYS,
+ zone_keys, &nkeys);
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "zone_resigninc:find_zone_keys -> %s\n",
+ dns_result_totext(result));
+ goto failure;
+ }
+
+ isc_stdtime_get(&now);
+ inception = now - 3600; /* Allow for clock skew. */
+ soaexpire = now + dns_zone_getsigvalidityinterval(zone);
+ /*
+ * Spread out signatures over time if they happen to be
+ * clumped. We don't do this for each add_sigs() call as
+ * we still want some clustering to occur.
+ */
+ isc_random_get(&jitter);
+ expire = soaexpire - jitter % 3600;
+ stop = now + 5;
+
+ check_ksk = DNS_ZONE_OPTION(zone, DNS_ZONEOPT_UPDATECHECKKSK);
+ if (check_ksk)
+ check_ksk = ksk_sanity(db, version);
+
+ name = dns_fixedname_name(&fixed);
+ result = dns_db_getsigningtime(db, &rdataset, name);
+ if (result != ISC_R_SUCCESS && result != ISC_R_NOTFOUND) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "zone_resigninc:dns_db_getsigningtime -> %s\n",
+ dns_result_totext(result));
+ }
+
+ i = 0;
+ while (result == ISC_R_SUCCESS) {
+ resign = rdataset.resign;
+ covers = rdataset.covers;
+ /*
+ * Stop if we hit the SOA as that means we have walked the
+ * entire zone. The SOA record should always be the most
+ * recent signature.
+ */
+ /* XXXMPA increase number of RRsets signed pre call */
+ if (covers == dns_rdatatype_soa || i++ > zone->signatures ||
+ resign > stop) {
+ /*
+ * Ensure that we don't loop resigning the SOA.
+ */
+ if (covers == dns_rdatatype_soa)
+ dns_db_resigned(db, &rdataset, version);
+ dns_rdataset_disassociate(&rdataset);
+ break;
+ }
+
+ dns_db_resigned(db, &rdataset, version);
+ dns_rdataset_disassociate(&rdataset);
+
+ result = del_sigs(zone, db, version, name, covers, &sig_diff,
+ zone_keys, nkeys, now);
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "zone_resigninc:del_sigs -> %s\n",
+ dns_result_totext(result));
+ break;
+ }
+ result = add_sigs(db, version, name, covers, &sig_diff,
+ zone_keys, nkeys, zone->mctx, inception,
+ expire, check_ksk);
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "zone_resigninc:add_sigs -> %s\n",
+ dns_result_totext(result));
+ break;
+ }
+ result = dns_db_getsigningtime(db, &rdataset,
+ dns_fixedname_name(&fixed));
+ if (nkeys == 0 && result == ISC_R_NOTFOUND) {
+ result = ISC_R_SUCCESS;
+ break;
+ }
+ if (result != ISC_R_SUCCESS)
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "zone_resigninc:dns_db_getsigningtime -> %s\n",
+ dns_result_totext(result));
+ }
+
+ if (result != ISC_R_NOMORE && result != ISC_R_SUCCESS)
+ goto failure;
+
+ result = del_sigs(zone, db, version, &zone->origin, dns_rdatatype_soa,
+ &sig_diff, zone_keys, nkeys, now);
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "zone_resigninc:del_sigs -> %s\n",
+ dns_result_totext(result));
+ goto failure;
+ }
+
+ result = increment_soa_serial(db, version, &sig_diff, zone->mctx);
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "zone_resigninc:increment_soa_serial -> %s\n",
+ dns_result_totext(result));
+ goto failure;
+ }
+
+ /*
+ * Generate maximum life time signatures so that the above loop
+ * termination is sensible.
+ */
+ result = add_sigs(db, version, &zone->origin, dns_rdatatype_soa,
+ &sig_diff, zone_keys, nkeys, zone->mctx, inception,
+ soaexpire, check_ksk);
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "zone_resigninc:add_sigs -> %s\n",
+ dns_result_totext(result));
+ goto failure;
+ }
+
+ journalfile = dns_zone_getjournal(zone);
+ if (journalfile != NULL) {
+ dns_journal_t *journal = NULL;
+ result = dns_journal_open(zone->mctx, journalfile,
+ ISC_TRUE, &journal);
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "zone_resigninc:dns_journal_open -> %s\n",
+ dns_result_totext(result));
+ goto failure;
+ }
+
+ result = dns_journal_write_transaction(journal, &sig_diff);
+ dns_journal_destroy(&journal);
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "zone_resigninc:dns_journal_write_transaction -> %s\n",
+ dns_result_totext(result));
+ goto failure;
+ }
+ }
+
+ /*
+ * Everything has succeeded. Commit the changes.
+ */
+ dns_db_closeversion(db, &version, ISC_TRUE);
+
+ failure:
+ dns_diff_clear(&sig_diff);
+ for (i = 0; i < nkeys; i++)
+ dst_key_free(&zone_keys[i]);
+ if (version != NULL) {
+ dns_db_closeversion(zone->db, &version, ISC_FALSE);
+ dns_db_detach(&db);
+ } else if (db != NULL)
+ dns_db_detach(&db);
+ if (result == ISC_R_SUCCESS) {
+ set_resigntime(zone);
+ LOCK_ZONE(zone);
+ zone_needdump(zone, DNS_DUMP_DELAY);
+ DNS_ZONE_SETFLAG(zone, DNS_ZONEFLG_NEEDNOTIFY);
+ UNLOCK_ZONE(zone);
+ } else {
+ /*
+ * Something failed. Retry in 5 minutes.
+ */
+ isc_interval_t ival;
+ isc_interval_set(&ival, 300, 0);
+ isc_time_nowplusinterval(&zone->resigntime, &ival);
+ }
+}
+
+static isc_result_t
+next_active(dns_db_t *db, dns_dbversion_t *version, dns_name_t *oldname,
+ dns_name_t *newname, isc_boolean_t bottom)
+{
+ isc_result_t result;
+ dns_dbiterator_t *dbit = NULL;
+ dns_rdatasetiter_t *rdsit = NULL;
+ dns_dbnode_t *node = NULL;
+
+ CHECK(dns_db_createiterator(db, DNS_DB_NONSEC3, &dbit));
+ CHECK(dns_dbiterator_seek(dbit, oldname));
+ do {
+ result = dns_dbiterator_next(dbit);
+ if (result == ISC_R_NOMORE)
+ CHECK(dns_dbiterator_first(dbit));
+ CHECK(dns_dbiterator_current(dbit, &node, newname));
+ if (bottom && dns_name_issubdomain(newname, oldname) &&
+ !dns_name_equal(newname, oldname)) {
+ dns_db_detachnode(db, &node);
+ continue;
+ }
+ /*
+ * Is this node empty?
+ */
+ CHECK(dns_db_allrdatasets(db, node, version, 0, &rdsit));
+ result = dns_rdatasetiter_first(rdsit);
+ dns_db_detachnode(db, &node);
+ dns_rdatasetiter_destroy(&rdsit);
+ if (result != ISC_R_NOMORE)
+ break;
+ } while (1);
+ failure:
+ if (node != NULL)
+ dns_db_detachnode(db, &node);
+ if (dbit != NULL)
+ dns_dbiterator_destroy(&dbit);
+ return (result);
+}
+
+static void
+set_bit(unsigned char *array, unsigned int index) {
+ unsigned int shift, mask;
+
+ shift = 7 - (index % 8);
+ mask = 1 << shift;
+
+ array[index / 8] |= mask;
+}
+
+static isc_boolean_t
+signed_with_key(dns_db_t *db, dns_dbnode_t *node, dns_dbversion_t *version,
+ dns_rdatatype_t type, dst_key_t *key)
+{
+ isc_result_t result;
+ dns_rdataset_t rdataset;
+ dns_rdata_t rdata = DNS_RDATA_INIT;
+ dns_rdata_rrsig_t rrsig;
+
+ dns_rdataset_init(&rdataset);
+ result = dns_db_findrdataset(db, node, version, dns_rdatatype_rrsig,
+ type, 0, &rdataset, NULL);
+ if (result != ISC_R_SUCCESS)
+ return (ISC_FALSE);
+ for (result = dns_rdataset_first(&rdataset);
+ result == ISC_R_SUCCESS;
+ result = dns_rdataset_next(&rdataset)) {
+ dns_rdataset_current(&rdataset, &rdata);
+ result = dns_rdata_tostruct(&rdata, &rrsig, NULL);
+ INSIST(result == ISC_R_SUCCESS);
+ if (rrsig.algorithm == dst_key_alg(key) &&
+ rrsig.keyid == dst_key_id(key)) {
+ dns_rdataset_disassociate(&rdataset);
+ return (ISC_TRUE);
+ }
+ dns_rdata_reset(&rdata);
+ }
+ dns_rdataset_disassociate(&rdataset);
+ return (ISC_FALSE);
+}
+
+static isc_result_t
+add_nsec(dns_db_t *db, dns_dbversion_t *version, dns_name_t *name,
+ dns_dbnode_t *node, dns_ttl_t ttl, isc_boolean_t bottom,
+ dns_diff_t *diff)
+{
+ dns_fixedname_t fixed;
+ dns_name_t *next;
+ dns_rdata_t rdata = DNS_RDATA_INIT;
+ isc_result_t result;
+ unsigned char nsecbuffer[DNS_NSEC_BUFFERSIZE];
+
+ dns_fixedname_init(&fixed);
+ next = dns_fixedname_name(&fixed);
+
+ CHECK(next_active(db, version, name, next, bottom));
+ CHECK(dns_nsec_buildrdata(db, version, node, next, nsecbuffer,
+ &rdata));
+ if (dns_name_equal(dns_db_origin(db), name)) {
+ /*
+ * Set the OPT bit to indicate that this is a
+ * partially secure zone.
+ */
+ isc_region_t region;
+
+ dns_rdata_toregion(&rdata, &region);
+ dns_name_fromregion(next, &region);
+ isc_region_consume(&region, next->length);
+ INSIST(region.length > (2 + dns_rdatatype_opt / 8) &&
+ region.base[0] == 0 &&
+ region.base[1] > dns_rdatatype_opt / 8);
+ set_bit(region.base + 2, dns_rdatatype_opt);
+ }
+ CHECK(update_one_rr(db, version, diff, DNS_DIFFOP_ADD, name, ttl,
+ &rdata));
+ failure:
+ return (result);
+}
+
+static isc_result_t
+sign_a_node(dns_db_t *db, dns_name_t *name, dns_dbnode_t *node,
+ dns_dbversion_t *version, isc_boolean_t build_nsec3,
+ isc_boolean_t build_nsec, dst_key_t *key,
+ isc_stdtime_t inception, isc_stdtime_t expire,
+ unsigned int minimum, isc_boolean_t is_ksk,
+ isc_boolean_t *delegation, dns_diff_t *diff,
+ isc_int32_t *signatures, isc_mem_t *mctx)
+{
+ isc_result_t result;
+ dns_rdatasetiter_t *iterator = NULL;
+ dns_rdataset_t rdataset;
+ dns_rdata_t rdata = DNS_RDATA_INIT;
+ isc_buffer_t buffer;
+ unsigned char data[1024];
+ isc_boolean_t seen_soa, seen_ns, seen_rr, seen_dname, seen_nsec,
+ seen_nsec3, seen_ds;
+ isc_boolean_t bottom;
+
+ result = dns_db_allrdatasets(db, node, version, 0, &iterator);
+ if (result != ISC_R_SUCCESS) {
+ if (result == ISC_R_NOTFOUND)
+ result = ISC_R_SUCCESS;
+ return (result);
+ }
+ dns_rdataset_init(&rdataset);
+ isc_buffer_init(&buffer, data, sizeof(data));
+ seen_rr = seen_soa = seen_ns = seen_dname = seen_nsec =
+ seen_nsec3 = seen_ds = ISC_FALSE;
+ for (result = dns_rdatasetiter_first(iterator);
+ result == ISC_R_SUCCESS;
+ result = dns_rdatasetiter_next(iterator)) {
+ dns_rdatasetiter_current(iterator, &rdataset);
+ if (rdataset.type == dns_rdatatype_soa)
+ seen_soa = ISC_TRUE;
+ else if (rdataset.type == dns_rdatatype_ns)
+ seen_ns = ISC_TRUE;
+ else if (rdataset.type == dns_rdatatype_ds)
+ seen_ds = ISC_TRUE;
+ else if (rdataset.type == dns_rdatatype_dname)
+ seen_dname = ISC_TRUE;
+ else if (rdataset.type == dns_rdatatype_nsec)
+ seen_nsec = ISC_TRUE;
+ else if (rdataset.type == dns_rdatatype_nsec3)
+ seen_nsec3 = ISC_TRUE;
+ seen_rr = ISC_TRUE;
+ dns_rdataset_disassociate(&rdataset);
+ }
+ if (result != ISC_R_NOMORE)
+ goto failure;
+ if (seen_ns && !seen_soa)
+ *delegation = ISC_TRUE;
+ /*
+ * Going from insecure to NSEC3.
+ * Don't generate NSEC3 records for NSEC3 records.
+ */
+ if (build_nsec3 && !seen_nsec3 && seen_rr) {
+ isc_boolean_t unsecure = !seen_ds && seen_ns && !seen_soa;
+ CHECK(dns_nsec3_addnsec3s(db, version, name, minimum,
+ unsecure, diff));
+ (*signatures)--;
+ }
+ /*
+ * Going from insecure to NSEC.
+ * Don't generate NSEC records for NSEC3 records.
+ */
+ if (build_nsec && !seen_nsec3 && !seen_nsec && seen_rr) {
+ /* Build and add NSEC. */
+ bottom = (seen_ns && !seen_soa) || seen_dname;
+ CHECK(add_nsec(db, version, name, node, minimum, bottom, diff));
+ /* Count a NSEC generation as a signature generation. */
+ (*signatures)--;
+ }
+ result = dns_rdatasetiter_first(iterator);
+ while (result == ISC_R_SUCCESS) {
+ dns_rdatasetiter_current(iterator, &rdataset);
+ if (rdataset.type == dns_rdatatype_soa ||
+ rdataset.type == dns_rdatatype_rrsig)
+ goto next_rdataset;
+ if (is_ksk && rdataset.type != dns_rdatatype_dnskey)
+ goto next_rdataset;
+ if (*delegation &&
+ rdataset.type != dns_rdatatype_ds &&
+ rdataset.type != dns_rdatatype_nsec)
+ goto next_rdataset;
+ if (signed_with_key(db, node, version, rdataset.type, key))
+ goto next_rdataset;
+ /* Calculate the signature, creating a RRSIG RDATA. */
+ CHECK(dns_dnssec_sign(name, &rdataset, key, &inception,
+ &expire, mctx, &buffer, &rdata));
+ /* Update the database and journal with the RRSIG. */
+ /* XXX inefficient - will cause dataset merging */
+ CHECK(update_one_rr(db, version, diff, DNS_DIFFOP_ADDRESIGN,
+ name, rdataset.ttl, &rdata));
+ dns_rdata_reset(&rdata);
+ (*signatures)--;
+ next_rdataset:
+ dns_rdataset_disassociate(&rdataset);
+ result = dns_rdatasetiter_next(iterator);
+ }
+ if (result == ISC_R_NOMORE)
+ result = ISC_R_SUCCESS;
+ if (seen_dname)
+ *delegation = ISC_TRUE;
+failure:
+ if (dns_rdataset_isassociated(&rdataset))
+ dns_rdataset_disassociate(&rdataset);
+ if (iterator != NULL)
+ dns_rdatasetiter_destroy(&iterator);
+ return (result);
+}
+
+static isc_result_t
+updatesecure(dns_db_t *db, dns_dbversion_t *version, dns_name_t *name,
+ dns_ttl_t minimum, isc_boolean_t *secureupdated, dns_diff_t *diff)
+{
+ isc_result_t result;
+ dns_rdata_t rdata = DNS_RDATA_INIT;
+ unsigned char nsecbuffer[DNS_NSEC_BUFFERSIZE];
+ dns_rdataset_t rdataset;
+ dns_rdata_nsec_t nsec;
+ dns_dbnode_t *node = NULL;
+
+ /*
+ * Check to see if the OPT bit has already been cleared.
+ */
+ CHECK(dns_db_getoriginnode(db, &node));
+ dns_rdataset_init(&rdataset);
+ CHECK(dns_db_findrdataset(db, node, version, dns_rdatatype_nsec,
+ dns_rdatatype_none, 0, &rdataset, NULL));
+ CHECK(dns_rdataset_first(&rdataset));
+ dns_rdataset_current(&rdataset, &rdata);
+
+ /*
+ * Find the NEXT name for building the new record.
+ */
+ CHECK(dns_rdata_tostruct(&rdata, &nsec, NULL));
+
+ /*
+ * Delete the old NSEC record.
+ */
+ CHECK(update_one_rr(db, version, diff, DNS_DIFFOP_DEL, name, minimum,
+ &rdata));
+ dns_rdata_reset(&rdata);
+
+ /*
+ * Add the new NSEC record.
+ */
+ CHECK(dns_nsec_buildrdata(db, version, node, &nsec.next, nsecbuffer,
+ &rdata));
+ CHECK(update_one_rr(db, version, diff, DNS_DIFFOP_ADD, name, minimum,
+ &rdata));
+ dns_rdata_reset(&rdata);
+
+ if (secureupdated != NULL)
+ *secureupdated = ISC_TRUE;
+
+ failure:
+ if (node != NULL)
+ dns_db_detachnode(db, &node);
+ if (dns_rdataset_isassociated(&rdataset))
+ dns_rdataset_disassociate(&rdataset);
+ return (result);
+}
+
+static isc_result_t
+updatesignwithkey(dns_signing_t *signing, dns_dbversion_t *version,
+ dns_name_t *name, dns_rdatatype_t privatetype,
+ dns_diff_t *diff)
+{
+ isc_result_t result;
+ dns_dbnode_t *node = NULL;
+ dns_rdataset_t rdataset;
+ dns_rdata_t rdata = DNS_RDATA_INIT;
+ unsigned char data[5];
+ isc_boolean_t seen_done = ISC_FALSE;
+
+ dns_rdataset_init(&rdataset);
+ result = dns_db_getoriginnode(signing->db, &node);
+ if (result != ISC_R_SUCCESS)
+ goto failure;
+
+ result = dns_db_findrdataset(signing->db, node, version, privatetype,
+ dns_rdatatype_none, 0, &rdataset, NULL);
+ if (result == ISC_R_NOTFOUND) {
+ result = ISC_R_SUCCESS;
+ goto failure;
+ }
+ if (result != ISC_R_SUCCESS)
+ goto failure;
+ for (result = dns_rdataset_first(&rdataset);
+ result == ISC_R_SUCCESS;
+ result = dns_rdataset_next(&rdataset)) {
+ dns_rdataset_current(&rdataset, &rdata);
+ if (rdata.length != 5 ||
+ rdata.data[0] != signing->algorithm ||
+ rdata.data[1] != ((signing->keyid >> 8) & 0xff) ||
+ rdata.data[2] != (signing->keyid & 0xff)) {
+ dns_rdata_reset(&rdata);
+ continue;
+ }
+ if (!signing->delete && rdata.data[4] != 0)
+ seen_done = ISC_TRUE;
+ else
+ CHECK(update_one_rr(signing->db, version, diff,
+ DNS_DIFFOP_DEL, name, rdataset.ttl, &rdata));
+ dns_rdata_reset(&rdata);
+ }
+ if (result == ISC_R_NOMORE)
+ result = ISC_R_SUCCESS;
+ if (!signing->delete && !seen_done) {
+
+ data[0] = signing->algorithm;
+ data[1] = (signing->keyid >> 8) & 0xff;
+ data[2] = signing->keyid & 0xff;
+ data[3] = 0;
+ data[4] = 1;
+ rdata.length = sizeof(data);
+ rdata.data = data;
+ rdata.type = privatetype;
+ rdata.rdclass = dns_db_class(signing->db);
+ CHECK(update_one_rr(signing->db, version, diff, DNS_DIFFOP_ADD,
+ name, rdataset.ttl, &rdata));
+ }
+ failure:
+ if (dns_rdataset_isassociated(&rdataset))
+ dns_rdataset_disassociate(&rdataset);
+ if (node != NULL)
+ dns_db_detachnode(signing->db, &node);
+ return (result);
+}
+
+static isc_result_t
+fixup_nsec3param(dns_db_t *db, dns_dbversion_t *ver, dns_nsec3chain_t *chain,
+ isc_boolean_t active, dns_diff_t *diff)
+{
+ dns_dbnode_t *node = NULL;
+ dns_name_t *name = dns_db_origin(db);
+ dns_rdata_t rdata = DNS_RDATA_INIT;
+ dns_rdataset_t rdataset;
+ dns_rdata_nsec3param_t nsec3param;
+ isc_result_t result;
+ isc_buffer_t buffer;
+ unsigned char parambuf[DNS_NSEC3PARAM_BUFFERSIZE];
+ dns_ttl_t ttl = 0;
+
+ dns_rdataset_init(&rdataset);
+
+ result = dns_db_getoriginnode(db, &node);
+ RUNTIME_CHECK(result == ISC_R_SUCCESS);
+ result = dns_db_findrdataset(db, node, ver, dns_rdatatype_nsec3param,
+ 0, 0, &rdataset, NULL);
+ if (result == ISC_R_NOTFOUND)
+ goto add;
+ if (result != ISC_R_SUCCESS)
+ goto failure;
+
+ /*
+ * Preserve the existing ttl.
+ */
+ ttl = rdataset.ttl;
+
+ /*
+ * Delete all NSEC3PARAM records which match that in nsec3chain.
+ */
+ for (result = dns_rdataset_first(&rdataset);
+ result == ISC_R_SUCCESS;
+ result = dns_rdataset_next(&rdataset)) {
+
+ dns_rdataset_current(&rdataset, &rdata);
+ CHECK(dns_rdata_tostruct(&rdata, &nsec3param, NULL));
+
+ if (nsec3param.hash != chain->nsec3param.hash ||
+ (active && nsec3param.flags != 0) ||
+ nsec3param.iterations != chain->nsec3param.iterations ||
+ nsec3param.salt_length != chain->nsec3param.salt_length ||
+ memcmp(nsec3param.salt, chain->nsec3param.salt,
+ nsec3param.salt_length)) {
+ dns_rdata_reset(&rdata);
+ continue;
+ }
+
+ CHECK(update_one_rr(db, ver, diff, DNS_DIFFOP_DEL,
+ name, rdataset.ttl, &rdata));
+ dns_rdata_reset(&rdata);
+ }
+ if (result != ISC_R_NOMORE)
+ goto failure;
+
+ add:
+ if ((chain->nsec3param.flags & DNS_NSEC3FLAG_REMOVE) != 0) {
+ result = ISC_R_SUCCESS;
+ goto failure;
+ }
+
+ /*
+ * Add a NSEC3PARAM record which matches that in nsec3chain but
+ * with all flags bits cleared.
+ *
+ * Note: we do not clear chain->nsec3param.flags as this change
+ * may be reversed.
+ */
+ isc_buffer_init(&buffer, &parambuf, sizeof(parambuf));
+ CHECK(dns_rdata_fromstruct(&rdata, dns_db_class(db),
+ dns_rdatatype_nsec3param,
+ &chain->nsec3param, &buffer));
+ rdata.data[1] = 0; /* Clear flag bits. */
+ CHECK(update_one_rr(db, ver, diff, DNS_DIFFOP_ADD, name, ttl, &rdata));
+
+ failure:
+ dns_db_detachnode(db, &node);
+ if (dns_rdataset_isassociated(&rdataset))
+ dns_rdataset_disassociate(&rdataset);
+ return (result);
+}
+
+static isc_result_t
+delete_nsec(dns_db_t *db, dns_dbversion_t *ver, dns_dbnode_t *node,
+ dns_name_t *name, dns_diff_t *diff)
+{
+ dns_rdataset_t rdataset;
+ isc_result_t result;
+
+ dns_rdataset_init(&rdataset);
+
+ result = dns_db_findrdataset(db, node, ver, dns_rdatatype_nsec,
+ 0, 0, &rdataset, NULL);
+ if (result == ISC_R_NOTFOUND)
+ return (ISC_R_SUCCESS);
+ if (result != ISC_R_SUCCESS)
+ return (result);
+ for (result = dns_rdataset_first(&rdataset);
+ result == ISC_R_SUCCESS;
+ result = dns_rdataset_next(&rdataset)) {
+ dns_rdata_t rdata = DNS_RDATA_INIT;
+
+ dns_rdataset_current(&rdataset, &rdata);
+ CHECK(update_one_rr(db, ver, diff, DNS_DIFFOP_DEL, name,
+ rdataset.ttl, &rdata));
+ }
+ if (result == ISC_R_NOMORE)
+ result = ISC_R_SUCCESS;
+ failure:
+ dns_rdataset_disassociate(&rdataset);
+ return (result);
+}
+
+static isc_result_t
+deletematchingnsec3(dns_db_t *db, dns_dbversion_t *ver, dns_dbnode_t *node,
+ dns_name_t *name, const dns_rdata_nsec3param_t *param,
+ dns_diff_t *diff)
+{
+ dns_rdataset_t rdataset;
+ dns_rdata_nsec3_t nsec3;
+ isc_result_t result;
+
+ dns_rdataset_init(&rdataset);
+ result = dns_db_findrdataset(db, node, ver, dns_rdatatype_nsec3,
+ 0, 0, &rdataset, NULL);
+ if (result == ISC_R_NOTFOUND)
+ return (ISC_R_SUCCESS);
+ if (result != ISC_R_SUCCESS)
+ return (result);
+
+ for (result = dns_rdataset_first(&rdataset);
+ result == ISC_R_SUCCESS;
+ result = dns_rdataset_next(&rdataset)) {
+ dns_rdata_t rdata = DNS_RDATA_INIT;
+
+ dns_rdataset_current(&rdataset, &rdata);
+ CHECK(dns_rdata_tostruct(&rdata, &nsec3, NULL));
+ if (nsec3.hash != param->hash ||
+ nsec3.iterations != param->iterations ||
+ nsec3.salt_length != param->salt_length ||
+ memcmp(nsec3.salt, param->salt, nsec3.salt_length))
+ continue;
+ CHECK(update_one_rr(db, ver, diff, DNS_DIFFOP_DEL, name,
+ rdataset.ttl, &rdata));
+ }
+ if (result == ISC_R_NOMORE)
+ result = ISC_R_SUCCESS;
+ failure:
+ dns_rdataset_disassociate(&rdataset);
+ return (result);
+}
+
+static isc_result_t
+need_nsec_chain(dns_db_t *db, dns_dbversion_t *ver,
+ const dns_rdata_nsec3param_t *param,
+ isc_boolean_t *answer, isc_boolean_t *updatensec)
+{
+ dns_dbnode_t *node = NULL;
+ dns_rdata_t rdata = DNS_RDATA_INIT;
+ dns_rdata_nsec3param_t myparam;
+ dns_rdataset_t rdataset;
+ isc_result_t result;
+
+ *answer = ISC_FALSE;
+
+ result = dns_db_getoriginnode(db, &node);
+ RUNTIME_CHECK(result == ISC_R_SUCCESS);
+
+ dns_rdataset_init(&rdataset);
+ result = dns_db_findrdataset(db, node, ver, dns_rdatatype_nsec,
+ 0, 0, &rdataset, NULL);
+ if (result == ISC_R_NOTFOUND)
+ goto check_nsec3param;
+
+ if (result != ISC_R_SUCCESS)
+ goto failure;
+
+ CHECK(dns_rdataset_first(&rdataset));
+ dns_rdataset_current(&rdataset, &rdata);
+
+ if (!dns_nsec_typepresent(&rdata, dns_rdatatype_opt)) {
+ /*
+ * We have a complete NSEC chain. Signal to update
+ * the apex NSEC record.
+ */
+ *updatensec = ISC_TRUE;
+ goto failure;
+ }
+ dns_rdataset_disassociate(&rdataset);
+ dns_rdata_reset(&rdata);
+
+ check_nsec3param:
+ result = dns_db_findrdataset(db, node, ver, dns_rdatatype_nsec3param,
+ 0, 0, &rdataset, NULL);
+ if (result == ISC_R_NOTFOUND) {
+ *answer = ISC_TRUE;
+ dns_db_detachnode(db, &node);
+ return (ISC_R_SUCCESS);
+ }
+ if (result != ISC_R_SUCCESS) {
+ dns_db_detachnode(db, &node);
+ return (result);
+ }
+
+ for (result = dns_rdataset_first(&rdataset);
+ result == ISC_R_SUCCESS;
+ result = dns_rdataset_next(&rdataset)) {
+ dns_rdataset_current(&rdataset, &rdata);
+ CHECK(dns_rdata_tostruct(&rdata, &myparam, NULL));
+ dns_rdata_reset(&rdata);
+ /*
+ * Ignore any NSEC3PARAM removals.
+ */
+ if (NSEC3REMOVE(myparam.flags))
+ continue;
+ /*
+ * Ignore the chain that we are in the process of deleting.
+ */
+ if (myparam.hash == param->hash &&
+ myparam.iterations == param->iterations &&
+ myparam.salt_length == param->salt_length &&
+ !memcmp(myparam.salt, param->salt, myparam.salt_length))
+ continue;
+ /*
+ * Found an active NSEC3 chain.
+ */
+ break;
+ }
+ if (result == ISC_R_NOMORE) {
+ *answer = ISC_TRUE;
+ result = ISC_R_SUCCESS;
+ }
+
+ failure:
+ if (dns_rdataset_isassociated(&rdataset))
+ dns_rdataset_disassociate(&rdataset);
+ dns_db_detachnode(db, &node);
+ return (result);
+}
+
+/*
+ * Incrementally build and sign a new NSEC3 chain using the parameters
+ * requested.
+ */
+static void
+zone_nsec3chain(dns_zone_t *zone) {
+ const char *journalfile;
+ dns_db_t *db = NULL;
+ dns_dbnode_t *node = NULL;
+ dns_dbversion_t *version = NULL;
+ dns_diff_t sig_diff;
+ dns_diff_t nsec_diff;
+ dns_diff_t nsec3_diff;
+ dns_diff_t param_diff;
+ dns_fixedname_t fixed;
+ dns_fixedname_t nextfixed;
+ dns_name_t *name, *nextname;
+ dns_rdataset_t rdataset;
+ dns_nsec3chain_t *nsec3chain = NULL, *nextnsec3chain;
+ dns_nsec3chainlist_t cleanup;
+ dst_key_t *zone_keys[MAXZONEKEYS];
+ isc_int32_t signatures;
+ isc_boolean_t check_ksk, is_ksk;
+ isc_boolean_t delegation;
+ isc_boolean_t first;
+ isc_result_t result;
+ isc_stdtime_t now, inception, soaexpire, expire, stop;
+ isc_uint32_t jitter;
+ unsigned int i;
+ unsigned int nkeys = 0;
+ isc_uint32_t nodes;
+ isc_boolean_t unsecure = ISC_FALSE;
+ isc_boolean_t seen_soa, seen_ns, seen_dname, seen_ds;
+ isc_boolean_t seen_nsec, seen_nsec3, seen_rr;
+ dns_rdatasetiter_t *iterator = NULL;
+ dns_difftuple_t *tuple;
+ isc_boolean_t buildnsecchain;
+ isc_boolean_t updatensec = ISC_FALSE;
+
+ dns_rdataset_init(&rdataset);
+ dns_fixedname_init(&fixed);
+ name = dns_fixedname_name(&fixed);
+ dns_fixedname_init(&nextfixed);
+ nextname = dns_fixedname_name(&nextfixed);
+ dns_diff_init(zone->mctx, &param_diff);
+ dns_diff_init(zone->mctx, &nsec3_diff);
+ dns_diff_init(zone->mctx, &nsec_diff);
+ dns_diff_init(zone->mctx, &sig_diff);
+ sig_diff.resign = zone->sigresigninginterval;
+ ISC_LIST_INIT(cleanup);
+
+ /*
+ * Updates are disabled. Pause for 5 minutes.
+ */
+ if (zone->update_disabled) {
+ result = ISC_R_FAILURE;
+ goto failure;
+ }
+
+ ZONEDB_LOCK(&zone->dblock, isc_rwlocktype_read);
+ dns_db_attach(zone->db, &db);
+ ZONEDB_UNLOCK(&zone->dblock, isc_rwlocktype_read);
+
+ result = dns_db_newversion(db, &version);
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "zone_nsec3chain:dns_db_newversion -> %s\n",
+ dns_result_totext(result));
+ goto failure;
+ }
+
+ result = find_zone_keys(zone, db, version, zone->mctx,
+ MAXZONEKEYS, zone_keys, &nkeys);
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "zone_nsec3chain:find_zone_keys -> %s\n",
+ dns_result_totext(result));
+ goto failure;
+ }
+
+ isc_stdtime_get(&now);
+ inception = now - 3600; /* Allow for clock skew. */
+ soaexpire = now + dns_zone_getsigvalidityinterval(zone);
+
+ /*
+ * Spread out signatures over time if they happen to be
+ * clumped. We don't do this for each add_sigs() call as
+ * we still want some clustering to occur.
+ */
+ isc_random_get(&jitter);
+ expire = soaexpire - jitter % 3600;
+ stop = now + 5;
+
+ check_ksk = DNS_ZONE_OPTION(zone, DNS_ZONEOPT_UPDATECHECKKSK);
+ if (check_ksk)
+ check_ksk = ksk_sanity(db, version);
+
+ /*
+ * We keep pulling nodes off each iterator in turn until
+ * we have no more nodes to pull off or we reach the limits
+ * for this quantum.
+ */
+ nodes = zone->nodes;
+ signatures = zone->signatures;
+ LOCK_ZONE(zone);
+ nsec3chain = ISC_LIST_HEAD(zone->nsec3chain);
+ UNLOCK_ZONE(zone);
+ first = ISC_TRUE;
+
+ if (nsec3chain != NULL)
+ nsec3chain->save_delete_nsec = nsec3chain->delete_nsec;
+ /*
+ * Generate new NSEC3 chains first.
+ */
+ while (nsec3chain != NULL && nodes-- > 0 && signatures > 0) {
+ LOCK_ZONE(zone);
+ nextnsec3chain = ISC_LIST_NEXT(nsec3chain, link);
+
+ ZONEDB_LOCK(&zone->dblock, isc_rwlocktype_read);
+ if (nsec3chain->done || nsec3chain->db != zone->db) {
+ ISC_LIST_UNLINK(zone->nsec3chain, nsec3chain, link);
+ ISC_LIST_APPEND(cleanup, nsec3chain, link);
+ }
+ ZONEDB_UNLOCK(&zone->dblock, isc_rwlocktype_read);
+ UNLOCK_ZONE(zone);
+ if (ISC_LIST_TAIL(cleanup) == nsec3chain)
+ goto next_addchain;
+
+ /*
+ * Possible future db.
+ */
+ if (nsec3chain->db != db) {
+ goto next_addchain;
+ }
+
+ if (NSEC3REMOVE(nsec3chain->nsec3param.flags))
+ goto next_addchain;
+
+ is_ksk = ISC_FALSE;
+ delegation = ISC_FALSE;
+ dns_dbiterator_current(nsec3chain->dbiterator, &node, name);
+
+ if (nsec3chain->delete_nsec) {
+ delegation = ISC_FALSE;
+ dns_dbiterator_pause(nsec3chain->dbiterator);
+ CHECK(delete_nsec(db, version, node, name, &nsec_diff));
+ goto next_addnode;
+ }
+ /*
+ * On the first pass we need to check if the current node
+ * has not been obscured.
+ */
+ delegation = ISC_FALSE;
+ unsecure = ISC_FALSE;
+ if (first) {
+ dns_fixedname_t ffound;
+ dns_name_t *found;
+ dns_fixedname_init(&ffound);
+ found = dns_fixedname_name(&ffound);
+ result = dns_db_find(db, name, version,
+ dns_rdatatype_soa,
+ DNS_DBFIND_NOWILD, 0, NULL, found,
+ NULL, NULL);
+ if ((result == DNS_R_DELEGATION ||
+ result == DNS_R_DNAME) &&
+ !dns_name_equal(name, found)) {
+ /*
+ * Remember the obscuring name so that
+ * we skip all obscured names.
+ */
+ dns_name_copy(found, name, NULL);
+ delegation = ISC_TRUE;
+ goto next_addnode;
+ }
+ }
+
+ /*
+ * Check to see if this is a bottom of zone node.
+ */
+ result = dns_db_allrdatasets(db, node, version, 0, &iterator);
+ if (result == ISC_R_NOTFOUND) /* Empty node? */
+ goto next_addnode;
+ if (result != ISC_R_SUCCESS)
+ goto failure;
+
+ seen_soa = seen_ns = seen_dname = seen_ds = seen_nsec =
+ ISC_FALSE;
+ for (result = dns_rdatasetiter_first(iterator);
+ result == ISC_R_SUCCESS;
+ result = dns_rdatasetiter_next(iterator)) {
+ dns_rdatasetiter_current(iterator, &rdataset);
+ INSIST(rdataset.type != dns_rdatatype_nsec3);
+ if (rdataset.type == dns_rdatatype_soa)
+ seen_soa = ISC_TRUE;
+ else if (rdataset.type == dns_rdatatype_ns)
+ seen_ns = ISC_TRUE;
+ else if (rdataset.type == dns_rdatatype_dname)
+ seen_dname = ISC_TRUE;
+ else if (rdataset.type == dns_rdatatype_ds)
+ seen_ds = ISC_TRUE;
+ else if (rdataset.type == dns_rdatatype_nsec)
+ seen_nsec = ISC_TRUE;
+ dns_rdataset_disassociate(&rdataset);
+ }
+ dns_rdatasetiter_destroy(&iterator);
+ /*
+ * Is there a NSEC chain than needs to be cleaned up?
+ */
+ if (seen_nsec)
+ nsec3chain->seen_nsec = ISC_TRUE;
+ if (seen_ns && !seen_soa && !seen_ds)
+ unsecure = ISC_TRUE;
+ if ((seen_ns && !seen_soa) || seen_dname)
+ delegation = ISC_TRUE;
+
+ /*
+ * Process one node.
+ */
+ dns_dbiterator_pause(nsec3chain->dbiterator);
+ CHECK(dns_nsec3_addnsec3(db, version, name,
+ &nsec3chain->nsec3param,
+ zone->minimum, unsecure, &nsec3_diff));
+ /*
+ * Treat each call to dns_nsec3_addnsec3() as if it's cost is
+ * two signatures. Additionally there will, in general, be
+ * two signature generated below.
+ *
+ * If we are only changing the optout flag the cost is half
+ * that of the cost of generating a completely new chain.
+ */
+ signatures -= 4;
+
+ /*
+ * Go onto next node.
+ */
+ next_addnode:
+ first = ISC_FALSE;
+ dns_db_detachnode(db, &node);
+ do {
+ result = dns_dbiterator_next(nsec3chain->dbiterator);
+
+ if (result == ISC_R_NOMORE && nsec3chain->delete_nsec) {
+ CHECK(fixup_nsec3param(db, version, nsec3chain,
+ ISC_FALSE, &param_diff));
+ LOCK_ZONE(zone);
+ ISC_LIST_UNLINK(zone->nsec3chain, nsec3chain,
+ link);
+ UNLOCK_ZONE(zone);
+ ISC_LIST_APPEND(cleanup, nsec3chain, link);
+ goto next_addchain;
+ }
+ if (result == ISC_R_NOMORE) {
+ dns_dbiterator_pause(nsec3chain->dbiterator);
+ if (nsec3chain->seen_nsec) {
+ CHECK(fixup_nsec3param(db, version,
+ nsec3chain,
+ ISC_TRUE,
+ &param_diff));
+ nsec3chain->delete_nsec = ISC_TRUE;
+ goto same_addchain;
+ }
+ CHECK(fixup_nsec3param(db, version, nsec3chain,
+ ISC_FALSE, &param_diff));
+ LOCK_ZONE(zone);
+ ISC_LIST_UNLINK(zone->nsec3chain, nsec3chain,
+ link);
+ UNLOCK_ZONE(zone);
+ ISC_LIST_APPEND(cleanup, nsec3chain, link);
+ goto next_addchain;
+ } else if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "zone_nsec3chain:"
+ "dns_dbiterator_next -> %s\n",
+ dns_result_totext(result));
+ goto failure;
+ } else if (delegation) {
+ dns_dbiterator_current(nsec3chain->dbiterator,
+ &node, nextname);
+ dns_db_detachnode(db, &node);
+ if (!dns_name_issubdomain(nextname, name))
+ break;
+ } else
+ break;
+ } while (1);
+ continue;
+
+ same_addchain:
+ CHECK(dns_dbiterator_first(nsec3chain->dbiterator));
+ first = ISC_TRUE;
+ continue;
+
+ next_addchain:
+ dns_dbiterator_pause(nsec3chain->dbiterator);
+ nsec3chain = nextnsec3chain;
+ first = ISC_TRUE;
+ if (nsec3chain != NULL)
+ nsec3chain->save_delete_nsec = nsec3chain->delete_nsec;
+ }
+
+ /*
+ * Process removals.
+ */
+ LOCK_ZONE(zone);
+ nsec3chain = ISC_LIST_HEAD(zone->nsec3chain);
+ UNLOCK_ZONE(zone);
+ first = ISC_TRUE;
+ buildnsecchain = ISC_FALSE;
+ while (nsec3chain != NULL && nodes-- > 0 && signatures > 0) {
+ LOCK_ZONE(zone);
+ nextnsec3chain = ISC_LIST_NEXT(nsec3chain, link);
+ UNLOCK_ZONE(zone);
+
+ if (nsec3chain->db != db)
+ goto next_removechain;
+
+ if (!NSEC3REMOVE(nsec3chain->nsec3param.flags))
+ goto next_removechain;
+
+ /*
+ * Work out if we need to build a NSEC chain as a consequence
+ * of removing this NSEC3 chain.
+ */
+ if (first && !updatensec &&
+ (nsec3chain->nsec3param.flags & DNS_NSEC3FLAG_NONSEC) == 0)
+ CHECK(need_nsec_chain(db, version,
+ &nsec3chain->nsec3param,
+ &buildnsecchain, &updatensec));
+
+ dns_dbiterator_current(nsec3chain->dbiterator, &node, name);
+ delegation = ISC_FALSE;
+
+ if (!buildnsecchain) {
+ /*
+ * Delete the NSECPARAM record that matches this chain.
+ */
+ if (first)
+ CHECK(fixup_nsec3param(db, version, nsec3chain,
+ ISC_TRUE, &param_diff));
+
+ /*
+ * Delete the NSEC3 records.
+ */
+ CHECK(deletematchingnsec3(db, version, node, name,
+ &nsec3chain->nsec3param,
+ &nsec3_diff));
+ goto next_removenode;
+ }
+
+ if (first) {
+ dns_fixedname_t ffound;
+ dns_name_t *found;
+ dns_fixedname_init(&ffound);
+ found = dns_fixedname_name(&ffound);
+ result = dns_db_find(db, name, version,
+ dns_rdatatype_soa,
+ DNS_DBFIND_NOWILD, 0, NULL, found,
+ NULL, NULL);
+ if ((result == DNS_R_DELEGATION ||
+ result == DNS_R_DNAME) &&
+ !dns_name_equal(name, found)) {
+ /*
+ * Remember the obscuring name so that
+ * we skip all obscured names.
+ */
+ dns_name_copy(found, name, NULL);
+ delegation = ISC_TRUE;
+ goto next_removenode;
+ }
+ }
+
+ /*
+ * Check to see if this is a bottom of zone node.
+ */
+ result = dns_db_allrdatasets(db, node, version, 0, &iterator);
+ if (result == ISC_R_NOTFOUND) /* Empty node? */
+ goto next_removenode;
+ if (result != ISC_R_SUCCESS)
+ goto failure;
+
+ seen_soa = seen_ns = seen_dname = seen_nsec3 = seen_nsec =
+ seen_rr = ISC_FALSE;
+ for (result = dns_rdatasetiter_first(iterator);
+ result == ISC_R_SUCCESS;
+ result = dns_rdatasetiter_next(iterator)) {
+ dns_rdatasetiter_current(iterator, &rdataset);
+ if (rdataset.type == dns_rdatatype_soa)
+ seen_soa = ISC_TRUE;
+ else if (rdataset.type == dns_rdatatype_ns)
+ seen_ns = ISC_TRUE;
+ else if (rdataset.type == dns_rdatatype_dname)
+ seen_dname = ISC_TRUE;
+ else if (rdataset.type == dns_rdatatype_nsec)
+ seen_nsec = ISC_TRUE;
+ else if (rdataset.type == dns_rdatatype_nsec3)
+ seen_nsec3 = ISC_TRUE;
+ seen_rr = ISC_TRUE;
+ dns_rdataset_disassociate(&rdataset);
+ }
+ dns_rdatasetiter_destroy(&iterator);
+
+ if (!seen_rr || seen_nsec3 || seen_nsec)
+ goto next_removenode;
+ if ((seen_ns && !seen_soa) || seen_dname)
+ delegation = ISC_TRUE;
+
+ CHECK(add_nsec(db, version, name, node, zone->minimum,
+ delegation, &nsec_diff));
+
+ next_removenode:
+ first = ISC_FALSE;
+ dns_db_detachnode(db, &node);
+ do {
+ result = dns_dbiterator_next(nsec3chain->dbiterator);
+ if (result == ISC_R_NOMORE && buildnsecchain) {
+ /*
+ * The NSEC chain should now be built.
+ * We can now remove the NSEC3 chain.
+ */
+ updatensec = ISC_TRUE;
+ goto same_removechain;
+ }
+ if (result == ISC_R_NOMORE) {
+ LOCK_ZONE(zone);
+ ISC_LIST_UNLINK(zone->nsec3chain, nsec3chain,
+ link);
+ UNLOCK_ZONE(zone);
+ ISC_LIST_APPEND(cleanup, nsec3chain, link);
+ dns_dbiterator_pause(nsec3chain->dbiterator);
+ CHECK(fixup_nsec3param(db, version, nsec3chain,
+ ISC_FALSE, &param_diff));
+ goto next_removechain;
+ } else if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "zone_nsec3chain:"
+ "dns_dbiterator_next -> %s\n",
+ dns_result_totext(result));
+ goto failure;
+ } else if (delegation) {
+ dns_dbiterator_current(nsec3chain->dbiterator,
+ &node, nextname);
+ dns_db_detachnode(db, &node);
+ if (!dns_name_issubdomain(nextname, name))
+ break;
+ } else
+ break;
+ } while (1);
+ continue;
+
+ same_removechain:
+ CHECK(dns_dbiterator_first(nsec3chain->dbiterator));
+ buildnsecchain = ISC_FALSE;
+ first = ISC_TRUE;
+ continue;
+
+ next_removechain:
+ dns_dbiterator_pause(nsec3chain->dbiterator);
+ nsec3chain = nextnsec3chain;
+ first = ISC_TRUE;
+ }
+
+ /*
+ * Add / update signatures for the NSEC3 records.
+ */
+ for (tuple = ISC_LIST_HEAD(nsec3_diff.tuples);
+ tuple != NULL;
+ tuple = ISC_LIST_HEAD(nsec3_diff.tuples)) {
+ /*
+ * We have changed the NSEC3 RRset above so we need to update
+ * the signatures.
+ */
+ result = del_sigs(zone, db, version, &tuple->name,
+ dns_rdatatype_nsec3, &sig_diff,
+ zone_keys, nkeys, now);
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "zone_nsec3chain:del_sigs -> %s\n",
+ dns_result_totext(result));
+ goto failure;
+ }
+ result = add_sigs(db, version, &tuple->name,
+ dns_rdatatype_nsec3, &sig_diff, zone_keys,
+ nkeys, zone->mctx, inception, expire,
+ check_ksk);
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "zone_nsec3chain:add_sigs -> %s\n",
+ dns_result_totext(result));
+ goto failure;
+ }
+
+ do {
+ dns_difftuple_t *next = ISC_LIST_NEXT(tuple, link);
+ while (next != NULL &&
+ !dns_name_equal(&tuple->name, &next->name))
+ next = ISC_LIST_NEXT(next, link);
+ ISC_LIST_UNLINK(nsec3_diff.tuples, tuple, link);
+ dns_diff_appendminimal(&sig_diff, &tuple);
+ INSIST(tuple == NULL);
+ tuple = next;
+ } while (tuple != NULL);
+ }
+
+ for (tuple = ISC_LIST_HEAD(param_diff.tuples);
+ tuple != NULL;
+ tuple = ISC_LIST_HEAD(param_diff.tuples)) {
+ /*
+ * We have changed the NSEC3PARAM RRset above so we need to
+ * update the signatures.
+ */
+ result = del_sigs(zone, db, version, &tuple->name,
+ dns_rdatatype_nsec3param, &sig_diff,
+ zone_keys, nkeys, now);
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "zone_nsec3chain:del_sigs -> %s\n",
+ dns_result_totext(result));
+ goto failure;
+ }
+ result = add_sigs(db, version, &tuple->name,
+ dns_rdatatype_nsec3param, &sig_diff,
+ zone_keys, nkeys, zone->mctx, inception,
+ expire, check_ksk);
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "zone_nsec3chain:add_sigs -> %s\n",
+ dns_result_totext(result));
+ goto failure;
+ }
+ ISC_LIST_UNLINK(param_diff.tuples, tuple, link);
+ dns_diff_appendminimal(&sig_diff, &tuple);
+ INSIST(tuple == NULL);
+ }
+
+ if (updatensec)
+ CHECK(updatesecure(db, version, &zone->origin, zone->minimum,
+ NULL, &nsec_diff));
+
+ for (tuple = ISC_LIST_HEAD(nsec_diff.tuples);
+ tuple != NULL;
+ tuple = ISC_LIST_HEAD(nsec_diff.tuples)) {
+ result = del_sigs(zone, db, version, &tuple->name,
+ dns_rdatatype_nsec, &sig_diff,
+ zone_keys, nkeys, now);
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "zone_nsec3chain:del_sigs -> %s\n",
+ dns_result_totext(result));
+ goto failure;
+ }
+ result = add_sigs(db, version, &tuple->name,
+ dns_rdatatype_nsec, &sig_diff,
+ zone_keys, nkeys, zone->mctx, inception,
+ expire, check_ksk);
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "zone_nsec3chain:add_sigs -> %s\n",
+ dns_result_totext(result));
+ goto failure;
+ }
+ ISC_LIST_UNLINK(nsec_diff.tuples, tuple, link);
+ dns_diff_appendminimal(&sig_diff, &tuple);
+ INSIST(tuple == NULL);
+ }
+
+ /*
+ * If we made no effective changes to the zone then we can just
+ * cleanup otherwise we need to increment the serial.
+ */
+ if (ISC_LIST_HEAD(sig_diff.tuples) == NULL)
+ goto done;
+
+ result = del_sigs(zone, db, version, &zone->origin, dns_rdatatype_soa,
+ &sig_diff, zone_keys, nkeys, now);
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR, "zone_nsec3chain:"
+ "del_sigs -> %s\n", dns_result_totext(result));
+ goto failure;
+ }
+
+ result = increment_soa_serial(db, version, &sig_diff, zone->mctx);
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR, "zone_nsec3chain:"
+ "increment_soa_serial -> %s\n",
+ dns_result_totext(result));
+ goto failure;
+ }
+
+ result = add_sigs(db, version, &zone->origin, dns_rdatatype_soa,
+ &sig_diff, zone_keys, nkeys, zone->mctx, inception,
+ soaexpire, check_ksk);
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR, "zone_nsec3chain:"
+ "add_sigs -> %s\n", dns_result_totext(result));
+ goto failure;
+ }
+
+ journalfile = dns_zone_getjournal(zone);
+ if (journalfile != NULL) {
+ dns_journal_t *journal = NULL;
+ result = dns_journal_open(zone->mctx, journalfile,
+ ISC_TRUE, &journal);
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR, "zone_nsec3chain:"
+ "dns_journal_open -> %s\n",
+ dns_result_totext(result));
+ goto failure;
+ }
+
+ result = dns_journal_write_transaction(journal, &sig_diff);
+ dns_journal_destroy(&journal);
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR, "zone_nsec3chain:"
+ "dns_journal_write_transaction -> %s\n",
+ dns_result_totext(result));
+ goto failure;
+ }
+ }
+
+ LOCK_ZONE(zone);
+ zone_needdump(zone, DNS_DUMP_DELAY);
+ UNLOCK_ZONE(zone);
+
+ done:
+ /*
+ * Pause all iterators so that dns_db_closeversion() can succeed.
+ */
+ LOCK_ZONE(zone);
+ for (nsec3chain = ISC_LIST_HEAD(zone->nsec3chain);
+ nsec3chain != NULL;
+ nsec3chain = ISC_LIST_NEXT(nsec3chain, link))
+ dns_dbiterator_pause(nsec3chain->dbiterator);
+ UNLOCK_ZONE(zone);
+
+ /*
+ * Everything has succeeded. Commit the changes.
+ */
+ dns_db_closeversion(db, &version, ISC_TRUE);
+
+ /*
+ * Everything succeeded so we can clean these up now.
+ */
+ nsec3chain = ISC_LIST_HEAD(cleanup);
+ while (nsec3chain != NULL) {
+ ISC_LIST_UNLINK(cleanup, nsec3chain, link);
+ dns_db_detach(&nsec3chain->db);
+ dns_dbiterator_destroy(&nsec3chain->dbiterator);
+ isc_mem_put(zone->mctx, nsec3chain, sizeof *nsec3chain);
+ nsec3chain = ISC_LIST_HEAD(cleanup);
+ }
+
+ set_resigntime(zone);
+
+ failure:
+ /*
+ * On error roll back the current nsec3chain.
+ */
+ if (result != ISC_R_SUCCESS && nsec3chain != NULL) {
+ if (nsec3chain->done) {
+ dns_db_detach(&nsec3chain->db);
+ dns_dbiterator_destroy(&nsec3chain->dbiterator);
+ isc_mem_put(zone->mctx, nsec3chain, sizeof *nsec3chain);
+ } else {
+ result = dns_dbiterator_first(nsec3chain->dbiterator);
+ RUNTIME_CHECK(result == ISC_R_SUCCESS);
+ dns_dbiterator_pause(nsec3chain->dbiterator);
+ nsec3chain->delete_nsec = nsec3chain->save_delete_nsec;
+ }
+ }
+
+ /*
+ * Rollback the cleanup list.
+ */
+ nsec3chain = ISC_LIST_TAIL(cleanup);
+ while (nsec3chain != NULL) {
+ ISC_LIST_UNLINK(cleanup, nsec3chain, link);
+ if (nsec3chain->done) {
+ dns_db_detach(&nsec3chain->db);
+ dns_dbiterator_destroy(&nsec3chain->dbiterator);
+ isc_mem_put(zone->mctx, nsec3chain, sizeof *nsec3chain);
+ } else {
+ LOCK_ZONE(zone);
+ ISC_LIST_PREPEND(zone->nsec3chain, nsec3chain, link);
+ UNLOCK_ZONE(zone);
+ result = dns_dbiterator_first(nsec3chain->dbiterator);
+ RUNTIME_CHECK(result == ISC_R_SUCCESS);
+ dns_dbiterator_pause(nsec3chain->dbiterator);
+ nsec3chain->delete_nsec = nsec3chain->save_delete_nsec;
+ }
+ nsec3chain = ISC_LIST_TAIL(cleanup);
+ }
+
+ LOCK_ZONE(zone);
+ for (nsec3chain = ISC_LIST_HEAD(zone->nsec3chain);
+ nsec3chain != NULL;
+ nsec3chain = ISC_LIST_NEXT(nsec3chain, link))
+ dns_dbiterator_pause(nsec3chain->dbiterator);
+ UNLOCK_ZONE(zone);
+
+ dns_diff_clear(&param_diff);
+ dns_diff_clear(&nsec3_diff);
+ dns_diff_clear(&nsec_diff);
+ dns_diff_clear(&sig_diff);
+
+ if (iterator != NULL)
+ dns_rdatasetiter_destroy(&iterator);
+
+ for (i = 0; i < nkeys; i++)
+ dst_key_free(&zone_keys[i]);
+
+ if (version != NULL) {
+ dns_db_closeversion(db, &version, ISC_FALSE);
+ dns_db_detach(&db);
+ } else if (db != NULL)
+ dns_db_detach(&db);
+
+ LOCK_ZONE(zone);
+ if (ISC_LIST_HEAD(zone->nsec3chain) != NULL) {
+ isc_interval_t i;
+ if (zone->update_disabled || result != ISC_R_SUCCESS)
+ isc_interval_set(&i, 60, 0); /* 1 minute */
+ else
+ isc_interval_set(&i, 0, 10000000); /* 10 ms */
+ isc_time_nowplusinterval(&zone->nsec3chaintime, &i);
+ } else
+ isc_time_settoepoch(&zone->nsec3chaintime);
+ UNLOCK_ZONE(zone);
+}
+
+static isc_result_t
+del_sig(dns_db_t *db, dns_dbversion_t *version, dns_name_t *name,
+ dns_dbnode_t *node, unsigned int nkeys, dns_secalg_t algorithm,
+ isc_uint16_t keyid, dns_diff_t *diff)
+{
+ dns_rdata_rrsig_t rrsig;
+ dns_rdataset_t rdataset;
+ dns_rdatasetiter_t *iterator = NULL;
+ isc_result_t result;
+
+ result = dns_db_allrdatasets(db, node, version, 0, &iterator);
+ if (result != ISC_R_SUCCESS) {
+ if (result == ISC_R_NOTFOUND)
+ result = ISC_R_SUCCESS;
+ return (result);
+ }
+
+ dns_rdataset_init(&rdataset);
+ for (result = dns_rdatasetiter_first(iterator);
+ result == ISC_R_SUCCESS;
+ result = dns_rdatasetiter_next(iterator)) {
+ dns_rdatasetiter_current(iterator, &rdataset);
+ if (nkeys == 0 && rdataset.type == dns_rdatatype_nsec) {
+ for (result = dns_rdataset_first(&rdataset);
+ result == ISC_R_SUCCESS;
+ result = dns_rdataset_next(&rdataset)) {
+ dns_rdata_t rdata = DNS_RDATA_INIT;
+ dns_rdataset_current(&rdataset, &rdata);
+ CHECK(update_one_rr(db, version, diff,
+ DNS_DIFFOP_DEL, name,
+ rdataset.ttl, &rdata));
+ }
+ if (result != ISC_R_NOMORE)
+ goto failure;
+ dns_rdataset_disassociate(&rdataset);
+ continue;
+ }
+ if (rdataset.type != dns_rdatatype_rrsig) {
+ dns_rdataset_disassociate(&rdataset);
+ continue;
+ }
+ for (result = dns_rdataset_first(&rdataset);
+ result == ISC_R_SUCCESS;
+ result = dns_rdataset_next(&rdataset)) {
+ dns_rdata_t rdata = DNS_RDATA_INIT;
+ dns_rdataset_current(&rdataset, &rdata);
+ CHECK(dns_rdata_tostruct(&rdata, &rrsig, NULL));
+ if (rrsig.algorithm != algorithm ||
+ rrsig.keyid != keyid)
+ continue;
+ CHECK(update_one_rr(db, version, diff,
+ DNS_DIFFOP_DEL, name,
+ rdataset.ttl, &rdata));
+ }
+ dns_rdataset_disassociate(&rdataset);
+ if (result != ISC_R_NOMORE)
+ break;
+ }
+ if (result == ISC_R_NOMORE)
+ result = ISC_R_SUCCESS;
+ failure:
+ if (dns_rdataset_isassociated(&rdataset))
+ dns_rdataset_disassociate(&rdataset);
+ dns_rdatasetiter_destroy(&iterator);
+ return (result);
+}
+
+/*
+ * Incrementally sign the zone using the keys requested.
+ * Builds the NSEC chain if required.
+ */
+static void
+zone_sign(dns_zone_t *zone) {
+ const char *journalfile;
+ dns_db_t *db = NULL;
+ dns_dbnode_t *node = NULL;
+ dns_dbversion_t *version = NULL;
+ dns_diff_t sig_diff;
+ dns_fixedname_t fixed;
+ dns_fixedname_t nextfixed;
+ dns_name_t *name, *nextname;
+ dns_rdataset_t rdataset;
+ dns_signing_t *signing, *nextsigning;
+ dns_signinglist_t cleanup;
+ dst_key_t *zone_keys[MAXZONEKEYS];
+ isc_int32_t signatures;
+ isc_boolean_t check_ksk, is_ksk;
+ isc_boolean_t delegation;
+ isc_boolean_t finishedakey = ISC_FALSE;
+ isc_boolean_t secureupdated = ISC_FALSE;
+ isc_boolean_t build_nsec3 = ISC_FALSE, build_nsec = ISC_FALSE;
+ isc_boolean_t first;
+ isc_result_t result;
+ isc_stdtime_t now, inception, soaexpire, expire, stop;
+ isc_uint32_t jitter;
+ unsigned int i;
+ unsigned int nkeys = 0;
+ isc_uint32_t nodes;
+
+ dns_rdataset_init(&rdataset);
+ dns_fixedname_init(&fixed);
+ name = dns_fixedname_name(&fixed);
+ dns_fixedname_init(&nextfixed);
+ nextname = dns_fixedname_name(&nextfixed);
+ dns_diff_init(zone->mctx, &sig_diff);
+ sig_diff.resign = zone->sigresigninginterval;
+ ISC_LIST_INIT(cleanup);
+
+ /*
+ * Updates are disabled. Pause for 5 minutes.
+ */
+ if (zone->update_disabled) {
+ result = ISC_R_FAILURE;
+ goto failure;
+ }
+
+ ZONEDB_LOCK(&zone->dblock, isc_rwlocktype_read);
+ dns_db_attach(zone->db, &db);
+ ZONEDB_UNLOCK(&zone->dblock, isc_rwlocktype_read);
+
+ result = dns_db_newversion(db, &version);
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "zone_sign:dns_db_newversion -> %s\n",
+ dns_result_totext(result));
+ goto failure;
+ }
+
+ result = find_zone_keys(zone, db, version, zone->mctx,
+ MAXZONEKEYS, zone_keys, &nkeys);
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "zone_sign:find_zone_keys -> %s\n",
+ dns_result_totext(result));
+ goto failure;
+ }
+
+ isc_stdtime_get(&now);
+ inception = now - 3600; /* Allow for clock skew. */
+ soaexpire = now + dns_zone_getsigvalidityinterval(zone);
+
+ /*
+ * Spread out signatures over time if they happen to be
+ * clumped. We don't do this for each add_sigs() call as
+ * we still want some clustering to occur.
+ */
+ isc_random_get(&jitter);
+ expire = soaexpire - jitter % 3600;
+ stop = now + 5;
+
+ check_ksk = DNS_ZONE_OPTION(zone, DNS_ZONEOPT_UPDATECHECKKSK);
+ if (check_ksk)
+ check_ksk = ksk_sanity(db, version);
+
+ /*
+ * We keep pulling nodes off each iterator in turn until
+ * we have no more nodes to pull off or we reach the limits
+ * for this quantum.
+ */
+ nodes = zone->nodes;
+ signatures = zone->signatures;
+ signing = ISC_LIST_HEAD(zone->signing);
+ first = ISC_TRUE;
+ /*
+ * See if we have a NSEC chain.
+ */
+ result = dns_db_getoriginnode(db, &node);
+ RUNTIME_CHECK(result == ISC_R_SUCCESS);
+ result = dns_db_findrdataset(db, node, version, dns_rdatatype_nsec,
+ dns_rdatatype_none, 0, &rdataset, NULL);
+ dns_db_detachnode(db, &node);
+ if (result == ISC_R_SUCCESS) {
+ build_nsec = ISC_TRUE;
+ dns_rdataset_disassociate(&rdataset);
+ } else if (result != ISC_R_NOTFOUND) {
+ goto failure;
+ } else {
+ /*
+ * No NSEC chain present.
+ * See if we need to build a NSEC3 chain?
+ */
+ result = dns_nsec3_active(db, version, ISC_TRUE, &build_nsec3);
+ if (result == ISC_R_SUCCESS) {
+ if (build_nsec3)
+ build_nsec3 = ISC_FALSE;
+ else {
+ result = dns_nsec3_active(db, version,
+ ISC_FALSE,
+ &build_nsec3);
+ if (build_nsec3)
+ secureupdated = ISC_TRUE;
+ else
+ build_nsec = ISC_TRUE;
+ }
+ }
+ }
+
+ while (signing != NULL && nodes-- > 0 && signatures > 0) {
+ nextsigning = ISC_LIST_NEXT(signing, link);
+
+ ZONEDB_LOCK(&zone->dblock, isc_rwlocktype_read);
+ if (signing->done || signing->db != zone->db) {
+ /*
+ * The zone has been reloaded. We will have
+ * created new signings as part of the reload
+ * process so we can destroy this one.
+ */
+ ISC_LIST_UNLINK(zone->signing, signing, link);
+ ISC_LIST_APPEND(cleanup, signing, link);
+ ZONEDB_UNLOCK(&zone->dblock, isc_rwlocktype_read);
+ goto next_signing;
+ }
+ ZONEDB_UNLOCK(&zone->dblock, isc_rwlocktype_read);
+
+ if (signing->db != db)
+ goto next_signing;
+
+ is_ksk = ISC_FALSE;
+ delegation = ISC_FALSE;
+
+ dns_dbiterator_current(signing->dbiterator, &node, name);
+
+ if (signing->delete) {
+ dns_dbiterator_pause(signing->dbiterator);
+ CHECK(del_sig(db, version, name, node, nkeys,
+ signing->algorithm, signing->keyid,
+ &sig_diff));
+ goto next_node;
+ }
+ /*
+ * On the first pass we need to check if the current node
+ * has not been obscured.
+ */
+ if (first) {
+ dns_fixedname_t ffound;
+ dns_name_t *found;
+ dns_fixedname_init(&ffound);
+ found = dns_fixedname_name(&ffound);
+ result = dns_db_find(db, name, version,
+ dns_rdatatype_soa,
+ DNS_DBFIND_NOWILD, 0, NULL, found,
+ NULL, NULL);
+ if ((result == DNS_R_DELEGATION ||
+ result == DNS_R_DNAME) &&
+ !dns_name_equal(name, found)) {
+ /*
+ * Remember the obscuring name so that
+ * we skip all obscured names.
+ */
+ dns_name_copy(found, name, NULL);
+ delegation = ISC_TRUE;
+ goto next_node;
+ }
+ }
+
+ /*
+ * Process one node.
+ */
+ dns_dbiterator_pause(signing->dbiterator);
+ for (i = 0; i < nkeys; i++) {
+ /*
+ * Find the key we want to sign with.
+ */
+ if (dst_key_alg(zone_keys[i]) != signing->algorithm ||
+ dst_key_id(zone_keys[i]) != signing->keyid ||
+ !dst_key_isprivate(zone_keys[i]))
+ continue;
+ /*
+ * Do we do KSK processing?
+ */
+ if (check_ksk &&
+ (dst_key_flags(zone_keys[i]) & DNS_KEYFLAG_KSK) != 0)
+ is_ksk = ISC_TRUE;
+ CHECK(sign_a_node(db, name, node, version, build_nsec3,
+ build_nsec, zone_keys[i], inception,
+ expire, zone->minimum, is_ksk,
+ &delegation, &sig_diff, &signatures,
+ zone->mctx));
+ break;
+ }
+ /*
+ * Go onto next node.
+ */
+ next_node:
+ first = ISC_FALSE;
+ dns_db_detachnode(db, &node);
+ do {
+ result = dns_dbiterator_next(signing->dbiterator);
+ if (result == ISC_R_NOMORE) {
+ ISC_LIST_UNLINK(zone->signing, signing, link);
+ ISC_LIST_APPEND(cleanup, signing, link);
+ dns_dbiterator_pause(signing->dbiterator);
+ finishedakey = ISC_TRUE;
+ if (!is_ksk && !secureupdated && nkeys != 0 &&
+ build_nsec) {
+ /*
+ * We have finished regenerating the
+ * zone with a zone signing key.
+ * The NSEC chain is now complete and
+ * there is a full set of signatures
+ * for the zone. We can now clear the
+ * OPT bit from the NSEC record.
+ */
+ result = updatesecure(db, version,
+ &zone->origin,
+ zone->minimum,
+ &secureupdated,
+ &sig_diff);
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone,
+ ISC_LOG_ERROR,
+ "updatesecure -> %s\n",
+ dns_result_totext(result));
+ goto failure;
+ }
+ }
+ result = updatesignwithkey(signing, version,
+ &zone->origin,
+ zone->privatetype,
+ &sig_diff);
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "updatesignwithkey -> %s\n",
+ dns_result_totext(result));
+ goto failure;
+ }
+ goto next_signing;
+ } else if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "zone_sign:dns_dbiterator_next -> %s\n",
+ dns_result_totext(result));
+ goto failure;
+ } else if (delegation) {
+ dns_dbiterator_current(signing->dbiterator,
+ &node, nextname);
+ dns_db_detachnode(db, &node);
+ if (!dns_name_issubdomain(nextname, name))
+ break;
+ } else
+ break;
+ } while (1);
+ continue;
+
+ next_signing:
+ dns_dbiterator_pause(signing->dbiterator);
+ signing = nextsigning;
+ first = ISC_TRUE;
+ }
+
+ if (secureupdated) {
+ /*
+ * We have changed the NSEC RRset above so we need to update
+ * the signatures.
+ */
+ result = del_sigs(zone, db, version, &zone->origin,
+ dns_rdatatype_nsec, &sig_diff, zone_keys,
+ nkeys, now);
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "zone_sign:del_sigs -> %s\n",
+ dns_result_totext(result));
+ goto failure;
+ }
+ result = add_sigs(db, version, &zone->origin,
+ dns_rdatatype_nsec, &sig_diff, zone_keys,
+ nkeys, zone->mctx, inception, soaexpire,
+ check_ksk);
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "zone_sign:add_sigs -> %s\n",
+ dns_result_totext(result));
+ goto failure;
+ }
+ }
+ if (finishedakey) {
+ /*
+ * We have changed the RRset above so we need to update
+ * the signatures.
+ */
+ result = del_sigs(zone, db, version, &zone->origin,
+ zone->privatetype, &sig_diff,
+ zone_keys, nkeys, now);
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "zone_sign:del_sigs -> %s\n",
+ dns_result_totext(result));
+ goto failure;
+ }
+ result = add_sigs(db, version, &zone->origin,
+ zone->privatetype, &sig_diff,
+ zone_keys, nkeys, zone->mctx, inception,
+ soaexpire, check_ksk);
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "zone_sign:add_sigs -> %s\n",
+ dns_result_totext(result));
+ goto failure;
+ }
+ }
+ result = del_sigs(zone, db, version, &zone->origin, dns_rdatatype_soa,
+ &sig_diff, zone_keys, nkeys, now);
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "zone_sign:del_sigs -> %s\n",
+ dns_result_totext(result));
+ goto failure;
+ }
+
+ result = increment_soa_serial(db, version, &sig_diff, zone->mctx);
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "zone_sign:increment_soa_serial -> %s\n",
+ dns_result_totext(result));
+ goto failure;
+ }
+
+ /*
+ * Generate maximum life time signatures so that the above loop
+ * termination is sensible.
+ */
+ result = add_sigs(db, version, &zone->origin, dns_rdatatype_soa,
+ &sig_diff, zone_keys, nkeys, zone->mctx, inception,
+ soaexpire, check_ksk);
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "zone_sign:add_sigs -> %s\n",
+ dns_result_totext(result));
+ goto failure;
+ }
+
+ journalfile = dns_zone_getjournal(zone);
+ if (journalfile != NULL) {
+ dns_journal_t *journal = NULL;
+ result = dns_journal_open(zone->mctx, journalfile,
+ ISC_TRUE, &journal);
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "zone_sign:dns_journal_open -> %s\n",
+ dns_result_totext(result));
+ goto failure;
+ }
+
+ result = dns_journal_write_transaction(journal, &sig_diff);
+ dns_journal_destroy(&journal);
+ if (result != ISC_R_SUCCESS) {
+ dns_zone_log(zone, ISC_LOG_ERROR,
+ "zone_sign:dns_journal_write_transaction -> %s\n",
+ dns_result_totext(result));
+ goto failure;
+ }
+ }
+
+
+ /*
+ * Pause all iterators so that dns_db_closeversion() can succeed.
+ */
+ for (signing = ISC_LIST_HEAD(zone->signing);
+ signing != NULL;
+ signing = ISC_LIST_NEXT(signing, link))
+ dns_dbiterator_pause(signing->dbiterator);
+
+ for (signing = ISC_LIST_HEAD(cleanup);
+ signing != NULL;
+ signing = ISC_LIST_NEXT(signing, link))
+ dns_dbiterator_pause(signing->dbiterator);
+
+ /*
+ * Everything has succeeded. Commit the changes.
+ */
+ dns_db_closeversion(db, &version, ISC_TRUE);
+
+ /*
+ * Everything succeeded so we can clean these up now.
+ */
+ signing = ISC_LIST_HEAD(cleanup);
+ while (signing != NULL) {
+ ISC_LIST_UNLINK(cleanup, signing, link);
+ dns_db_detach(&signing->db);
+ dns_dbiterator_destroy(&signing->dbiterator);
+ isc_mem_put(zone->mctx, signing, sizeof *signing);
+ signing = ISC_LIST_HEAD(cleanup);
+ }
+
+ set_resigntime(zone);
+
+ LOCK_ZONE(zone);
+ zone_needdump(zone, DNS_DUMP_DELAY);
+ UNLOCK_ZONE(zone);
+
+ failure:
+ /*
+ * Rollback the cleanup list.
+ */
+ signing = ISC_LIST_HEAD(cleanup);
+ while (signing != NULL) {
+ ISC_LIST_UNLINK(cleanup, signing, link);
+ ISC_LIST_APPEND(zone->signing, signing, link);
+ dns_dbiterator_first(signing->dbiterator);
+ dns_dbiterator_pause(signing->dbiterator);
+ signing = ISC_LIST_HEAD(cleanup);
+ }
+
+ for (signing = ISC_LIST_HEAD(zone->signing);
+ signing != NULL;
+ signing = ISC_LIST_NEXT(signing, link))
+ dns_dbiterator_pause(signing->dbiterator);
+
+ dns_diff_clear(&sig_diff);
+
+ for (i = 0; i < nkeys; i++)
+ dst_key_free(&zone_keys[i]);
+
+ if (version != NULL) {
+ dns_db_closeversion(db, &version, ISC_FALSE);
+ dns_db_detach(&db);
+ } else if (db != NULL)
+ dns_db_detach(&db);
+
+ if (ISC_LIST_HEAD(zone->signing) != NULL) {
+ isc_interval_t i;
+ if (zone->update_disabled || result != ISC_R_SUCCESS)
+ isc_interval_set(&i, 60, 0); /* 1 minute */
+ else
+ isc_interval_set(&i, 0, 10000000); /* 10 ms */
+ isc_time_nowplusinterval(&zone->signingtime, &i);
+ } else
+ isc_time_settoepoch(&zone->signingtime);
+}
+
static void
zone_maintenance(dns_zone_t *zone) {
const char me[] = "zone_maintenance";
@@ -3016,15 +5981,34 @@ zone_maintenance(dns_zone_t *zone) {
break;
}
- /*
- * Do we need to send out notify messages?
- */
switch (zone->type) {
case dns_zone_master:
case dns_zone_slave:
+ /*
+ * Do we need to send out notify messages?
+ */
if (DNS_ZONE_FLAG(zone, DNS_ZONEFLG_NEEDNOTIFY) &&
isc_time_compare(&now, &zone->notifytime) >= 0)
zone_notify(zone, &now);
+ /*
+ * Do we need to sign/resign some RRsets?
+ */
+ if (!isc_time_isepoch(&zone->signingtime) &&
+ isc_time_compare(&now, &zone->signingtime) >= 0)
+ zone_sign(zone);
+ else if (!isc_time_isepoch(&zone->resigntime) &&
+ isc_time_compare(&now, &zone->resigntime) >= 0)
+ zone_resigninc(zone);
+ else if (!isc_time_isepoch(&zone->nsec3chaintime) &&
+ isc_time_compare(&now, &zone->nsec3chaintime) >= 0)
+ zone_nsec3chain(zone);
+ /*
+ * Do we need to issue a key expiry warning.
+ */
+ if (!isc_time_isepoch(&zone->keywarntime) &&
+ isc_time_compare(&now, &zone->keywarntime) >= 0)
+ set_key_expiry_warning(zone, zone->key_expiry,
+ isc_time_seconds(&now));
break;
default:
break;
@@ -3036,6 +6020,7 @@ void
dns_zone_markdirty(dns_zone_t *zone) {
LOCK_ZONE(zone);
+ set_resigntime(zone); /* XXXMPA make separate call back */
zone_needdump(zone, DNS_DUMP_DELAY);
UNLOCK_ZONE(zone);
}
@@ -3776,6 +6761,16 @@ notify_send_toaddr(isc_task_t *task, isc_event_t *event) {
timeout * 3, timeout,
notify->zone->task, notify_done,
notify, &notify->request);
+ if (result == ISC_R_SUCCESS) {
+ if (isc_sockaddr_pf(&notify->dst) == AF_INET) {
+ inc_stats(notify->zone,
+ dns_zonestatscounter_notifyoutv4);
+ } else {
+ inc_stats(notify->zone,
+ dns_zonestatscounter_notifyoutv6);
+ }
+ }
+
cleanup_key:
if (key != NULL)
dns_tsigkey_detach(&key);
@@ -3975,9 +6970,11 @@ zone_notify(dns_zone_t *zone, isc_time_t *now) {
RUNTIME_CHECK(result == ISC_R_SUCCESS);
dns_rdata_reset(&rdata);
/*
- * don't notify the master server.
+ * Don't notify the master server unless explicitly
+ * configured to do so.
*/
- if (dns_name_compare(&master, &ns.name) == 0) {
+ if (!DNS_ZONE_OPTION(zone, DNS_ZONEOPT_NOTIFYTOSOA) &&
+ dns_name_compare(&master, &ns.name) == 0) {
result = dns_rdataset_next(&nsrdset);
continue;
}
@@ -4163,6 +7160,8 @@ stub_callback(isc_task_t *task, isc_event_t *event) {
master, source);
goto same_master;
}
+ dns_zonemgr_unreachableadd(zone->zmgr, &zone->masteraddr,
+ &zone->sourceaddr, &now);
dns_zone_log(zone, ISC_LOG_INFO,
"could not refresh stub from master %s"
" (source %s): %s", master, source,
@@ -4424,12 +7423,23 @@ refresh_callback(isc_task_t *task, isc_event_t *event) {
"master %s exceeded (source %s)",
master, source);
/* Try with slave with TCP. */
- if (zone->type == dns_zone_slave) {
- LOCK_ZONE(zone);
- DNS_ZONE_SETFLAG(zone,
- DNS_ZONEFLG_SOABEFOREAXFR);
- UNLOCK_ZONE(zone);
- goto tcp_transfer;
+ if (zone->type == dns_zone_slave &&
+ DNS_ZONE_OPTION(zone, DNS_ZONEOPT_TRYTCPREFRESH)) {
+ if (!dns_zonemgr_unreachable(zone->zmgr,
+ &zone->masteraddr,
+ &zone->sourceaddr,
+ &now)) {
+ LOCK_ZONE(zone);
+ DNS_ZONE_SETFLAG(zone,
+ DNS_ZONEFLG_SOABEFOREAXFR);
+ UNLOCK_ZONE(zone);
+ goto tcp_transfer;
+ }
+ dns_zone_log(zone, ISC_LOG_DEBUG(1),
+ "refresh: skipped tcp fallback"
+ "as master %s (source %s) is "
+ "unreachable (cached)",
+ master, source);
}
} else
dns_zone_log(zone, ISC_LOG_INFO,
@@ -4479,7 +7489,7 @@ refresh_callback(isc_task_t *task, isc_event_t *event) {
"master %s (source %s)", (int)rb.used, rcode,
master, source);
/*
- * Perhaps AXFR/IXFR is allowed even if SOA queries arn't.
+ * Perhaps AXFR/IXFR is allowed even if SOA queries aren't.
*/
if (msg->rcode == dns_rcode_refused &&
zone->type == dns_zone_slave)
@@ -4605,6 +7615,16 @@ refresh_callback(isc_task_t *task, isc_event_t *event) {
if (!DNS_ZONE_FLAG(zone, DNS_ZONEFLG_LOADED) ||
DNS_ZONE_FLAG(zone, DNS_ZONEFLG_FORCEXFER) ||
isc_serial_gt(serial, zone->serial)) {
+ if (dns_zonemgr_unreachable(zone->zmgr, &zone->masteraddr,
+ &zone->sourceaddr, &now)) {
+ dns_zone_log(zone, ISC_LOG_INFO,
+ "refresh: skipping %s as master %s "
+ "(source %s) is unreachable (cached)",
+ zone->type == dns_zone_slave ?
+ "zone transfer" : "NS query",
+ master, source);
+ goto next_master;
+ }
tcp_transfer:
isc_event_free(&event);
LOCK_ZONE(zone);
@@ -4820,7 +7840,7 @@ create_query(dns_zone_t *zone, dns_rdatatype_t rdtype,
}
static isc_result_t
-add_opt(dns_message_t *message, isc_uint16_t udpsize) {
+add_opt(dns_message_t *message, isc_uint16_t udpsize, isc_boolean_t reqnsid) {
dns_rdataset_t *rdataset = NULL;
dns_rdatalist_t *rdatalist = NULL;
dns_rdata_t *rdata = NULL;
@@ -4850,11 +7870,21 @@ add_opt(dns_message_t *message, isc_uint16_t udpsize) {
*/
rdatalist->ttl = 0;
- /*
- * No EDNS options.
- */
- rdata->data = NULL;
- rdata->length = 0;
+ /* Set EDNS options if applicable */
+ if (reqnsid) {
+ unsigned char data[4];
+ isc_buffer_t buf;
+
+ isc_buffer_init(&buf, data, sizeof(data));
+ isc_buffer_putuint16(&buf, DNS_OPT_NSID);
+ isc_buffer_putuint16(&buf, 0);
+ rdata->data = data;
+ rdata->length = sizeof(data);
+ } else {
+ rdata->data = NULL;
+ rdata->length = 0;
+ }
+
rdata->rdclass = rdatalist->rdclass;
rdata->type = rdatalist->type;
rdata->flags = 0;
@@ -4889,7 +7919,7 @@ soa_query(isc_task_t *task, isc_event_t *event) {
isc_uint32_t options;
isc_boolean_t cancel = ISC_TRUE;
int timeout;
- isc_boolean_t have_xfrsource;
+ isc_boolean_t have_xfrsource, reqnsid;
isc_uint16_t udpsize = SEND_BUFFER_SIZE;
REQUIRE(DNS_ZONE_VALID(zone));
@@ -4941,6 +7971,7 @@ soa_query(isc_task_t *task, isc_event_t *event) {
(void)dns_view_getpeertsig(zone->view, &masterip, &key);
have_xfrsource = ISC_FALSE;
+ reqnsid = zone->view->requestnsid;
if (zone->view->peers != NULL) {
dns_peer_t *peer = NULL;
isc_boolean_t edns;
@@ -4958,6 +7989,7 @@ soa_query(isc_task_t *task, isc_event_t *event) {
udpsize =
dns_resolver_getudpsize(zone->view->resolver);
(void)dns_peer_getudpsize(peer, &udpsize);
+ (void)dns_peer_getrequestnsid(peer, &reqnsid);
}
}
@@ -4989,7 +8021,7 @@ soa_query(isc_task_t *task, isc_event_t *event) {
DNS_REQUESTOPT_TCP : 0;
if (!DNS_ZONE_FLAG(zone, DNS_ZONEFLG_NOEDNS)) {
- result = add_opt(message, udpsize);
+ result = add_opt(message, udpsize, reqnsid);
if (result != ISC_R_SUCCESS)
zone_debuglog(zone, me, 1,
"unable to add opt record: %s",
@@ -5011,6 +8043,11 @@ soa_query(isc_task_t *task, isc_event_t *event) {
"dns_request_createvia2() failed: %s",
dns_result_totext(result));
goto cleanup;
+ } else {
+ if (isc_sockaddr_pf(&zone->masteraddr) == PF_INET)
+ inc_stats(zone, dns_zonestatscounter_soaoutv4);
+ else
+ inc_stats(zone, dns_zonestatscounter_soaoutv6);
}
cancel = ISC_FALSE;
@@ -5053,7 +8090,7 @@ ns_query(dns_zone_t *zone, dns_rdataset_t *soardataset, dns_stub_t *stub) {
dns_tsigkey_t *key = NULL;
dns_dbnode_t *node = NULL;
int timeout;
- isc_boolean_t have_xfrsource = ISC_FALSE;
+ isc_boolean_t have_xfrsource = ISC_FALSE, reqnsid;
isc_uint16_t udpsize = SEND_BUFFER_SIZE;
REQUIRE(DNS_ZONE_VALID(zone));
@@ -5165,6 +8202,7 @@ ns_query(dns_zone_t *zone, dns_rdataset_t *soardataset, dns_stub_t *stub) {
if (key == NULL)
(void)dns_view_getpeertsig(zone->view, &masterip, &key);
+ reqnsid = zone->view->requestnsid;
if (zone->view->peers != NULL) {
dns_peer_t *peer = NULL;
isc_boolean_t edns;
@@ -5182,11 +8220,12 @@ ns_query(dns_zone_t *zone, dns_rdataset_t *soardataset, dns_stub_t *stub) {
udpsize =
dns_resolver_getudpsize(zone->view->resolver);
(void)dns_peer_getudpsize(peer, &udpsize);
+ (void)dns_peer_getrequestnsid(peer, &reqnsid);
}
}
if (!DNS_ZONE_FLAG(zone, DNS_ZONEFLG_NOEDNS)) {
- result = add_opt(message, udpsize);
+ result = add_opt(message, udpsize, reqnsid);
if (result != ISC_R_SUCCESS)
zone_debuglog(zone, me, 1,
"unable to add opt record: %s",
@@ -5382,6 +8421,26 @@ zone_settimer(dns_zone_t *zone, isc_time_t *now) {
isc_time_compare(&zone->dumptime, &next) < 0)
next = zone->dumptime;
}
+ if (!isc_time_isepoch(&zone->resigntime)) {
+ if (isc_time_isepoch(&next) ||
+ isc_time_compare(&zone->resigntime, &next) < 0)
+ next = zone->resigntime;
+ }
+ if (!isc_time_isepoch(&zone->keywarntime)) {
+ if (isc_time_isepoch(&next) ||
+ isc_time_compare(&zone->keywarntime, &next) < 0)
+ next = zone->keywarntime;
+ }
+ if (!isc_time_isepoch(&zone->signingtime)) {
+ if (isc_time_isepoch(&next) ||
+ isc_time_compare(&zone->signingtime, &next) < 0)
+ next = zone->signingtime;
+ }
+ if (!isc_time_isepoch(&zone->nsec3chaintime)) {
+ if (isc_time_isepoch(&next) ||
+ isc_time_compare(&zone->nsec3chaintime, &next) < 0)
+ next = zone->nsec3chaintime;
+ }
break;
case dns_zone_slave:
@@ -5651,6 +8710,10 @@ dns_zone_notifyreceive(dns_zone_t *zone, isc_sockaddr_t *from,
* We only handle NOTIFY (SOA) at the present.
*/
LOCK_ZONE(zone);
+ if (isc_sockaddr_pf(from) == PF_INET)
+ inc_stats(zone, dns_zonestatscounter_notifyinv4);
+ else
+ inc_stats(zone, dns_zonestatscounter_notifyinv6);
if (msg->counts[DNS_SECTION_QUESTION] == 0 ||
dns_message_findname(msg, DNS_SECTION_QUESTION, &zone->origin,
dns_rdatatype_soa, dns_rdatatype_none,
@@ -5705,6 +8768,7 @@ dns_zone_notifyreceive(dns_zone_t *zone, isc_sockaddr_t *from,
UNLOCK_ZONE(zone);
dns_zone_log(zone, ISC_LOG_INFO,
"refused notify from non-master: %s", fromtext);
+ inc_stats(zone, dns_zonestatscounter_notifyrej);
return (DNS_R_REFUSED);
}
@@ -5788,6 +8852,18 @@ dns_zone_setqueryacl(dns_zone_t *zone, dns_acl_t *acl) {
}
void
+dns_zone_setqueryonacl(dns_zone_t *zone, dns_acl_t *acl) {
+
+ REQUIRE(DNS_ZONE_VALID(zone));
+
+ LOCK_ZONE(zone);
+ if (zone->queryon_acl != NULL)
+ dns_acl_detach(&zone->queryon_acl);
+ dns_acl_attach(acl, &zone->queryon_acl);
+ UNLOCK_ZONE(zone);
+}
+
+void
dns_zone_setupdateacl(dns_zone_t *zone, dns_acl_t *acl) {
REQUIRE(DNS_ZONE_VALID(zone));
@@ -5840,6 +8916,14 @@ dns_zone_getqueryacl(dns_zone_t *zone) {
}
dns_acl_t *
+dns_zone_getqueryonacl(dns_zone_t *zone) {
+
+ REQUIRE(DNS_ZONE_VALID(zone));
+
+ return (zone->queryon_acl);
+}
+
+dns_acl_t *
dns_zone_getupdateacl(dns_zone_t *zone) {
REQUIRE(DNS_ZONE_VALID(zone));
@@ -5908,6 +8992,17 @@ dns_zone_clearqueryacl(dns_zone_t *zone) {
}
void
+dns_zone_clearqueryonacl(dns_zone_t *zone) {
+
+ REQUIRE(DNS_ZONE_VALID(zone));
+
+ LOCK_ZONE(zone);
+ if (zone->queryon_acl != NULL)
+ dns_acl_detach(&zone->queryon_acl);
+ UNLOCK_ZONE(zone);
+}
+
+void
dns_zone_clearxfracl(dns_zone_t *zone) {
REQUIRE(DNS_ZONE_VALID(zone));
@@ -5977,7 +9072,7 @@ dns_zone_getjournalsize(dns_zone_t *zone) {
}
static void
-zone_tostr(dns_zone_t *zone, char *buf, size_t length) {
+zone_namerd_tostr(dns_zone_t *zone, char *buf, size_t length) {
isc_result_t result = ISC_R_FAILURE;
isc_buffer_t buffer;
@@ -6008,29 +9103,88 @@ zone_tostr(dns_zone_t *zone, char *buf, size_t length) {
buf[isc_buffer_usedlength(&buffer)] = '\0';
}
+static void
+zone_name_tostr(dns_zone_t *zone, char *buf, size_t length) {
+ isc_result_t result = ISC_R_FAILURE;
+ isc_buffer_t buffer;
+
+ REQUIRE(buf != NULL);
+ REQUIRE(length > 1U);
+
+ /*
+ * Leave space for terminating '\0'.
+ */
+ isc_buffer_init(&buffer, buf, length - 1);
+ if (dns_name_dynamic(&zone->origin))
+ result = dns_name_totext(&zone->origin, ISC_TRUE, &buffer);
+ if (result != ISC_R_SUCCESS &&
+ isc_buffer_availablelength(&buffer) >= (sizeof("<UNKNOWN>") - 1))
+ isc_buffer_putstr(&buffer, "<UNKNOWN>");
+
+ buf[isc_buffer_usedlength(&buffer)] = '\0';
+}
+
+static void
+zone_rdclass_tostr(dns_zone_t *zone, char *buf, size_t length) {
+ isc_buffer_t buffer;
+
+ REQUIRE(buf != NULL);
+ REQUIRE(length > 1U);
+
+ /*
+ * Leave space for terminating '\0'.
+ */
+ isc_buffer_init(&buffer, buf, length - 1);
+ (void)dns_rdataclass_totext(zone->rdclass, &buffer);
+
+ buf[isc_buffer_usedlength(&buffer)] = '\0';
+}
+
+static void
+zone_viewname_tostr(dns_zone_t *zone, char *buf, size_t length) {
+ isc_buffer_t buffer;
+
+ REQUIRE(buf != NULL);
+ REQUIRE(length > 1U);
+
+
+ /*
+ * Leave space for terminating '\0'.
+ */
+ isc_buffer_init(&buffer, buf, length - 1);
+
+ if (zone->view == NULL) {
+ isc_buffer_putstr(&buffer, "_none");
+ } else if (strlen(zone->view->name)
+ < isc_buffer_availablelength(&buffer)) {
+ isc_buffer_putstr(&buffer, zone->view->name);
+ } else {
+ isc_buffer_putstr(&buffer, "_toolong");
+ }
+
+ buf[isc_buffer_usedlength(&buffer)] = '\0';
+}
+
void
dns_zone_name(dns_zone_t *zone, char *buf, size_t length) {
REQUIRE(DNS_ZONE_VALID(zone));
REQUIRE(buf != NULL);
- zone_tostr(zone, buf, length);
+ zone_namerd_tostr(zone, buf, length);
}
static void
notify_log(dns_zone_t *zone, int level, const char *fmt, ...) {
va_list ap;
char message[4096];
- char namebuf[1024+32];
if (isc_log_wouldlog(dns_lctx, level) == ISC_FALSE)
return;
- zone_tostr(zone, namebuf, sizeof(namebuf));
-
va_start(ap, fmt);
vsnprintf(message, sizeof(message), fmt, ap);
va_end(ap);
isc_log_write(dns_lctx, DNS_LOGCATEGORY_NOTIFY, DNS_LOGMODULE_ZONE,
- level, "zone %s: %s", namebuf, message);
+ level, "zone %s: %s", zone->strnamerd, message);
}
void
@@ -6038,36 +9192,30 @@ dns_zone_logc(dns_zone_t *zone, isc_logcategory_t *category,
int level, const char *fmt, ...) {
va_list ap;
char message[4096];
- char namebuf[1024+32];
if (isc_log_wouldlog(dns_lctx, level) == ISC_FALSE)
return;
- zone_tostr(zone, namebuf, sizeof(namebuf));
-
va_start(ap, fmt);
vsnprintf(message, sizeof(message), fmt, ap);
va_end(ap);
isc_log_write(dns_lctx, category, DNS_LOGMODULE_ZONE,
- level, "zone %s: %s", namebuf, message);
+ level, "zone %s: %s", zone->strnamerd, message);
}
void
dns_zone_log(dns_zone_t *zone, int level, const char *fmt, ...) {
va_list ap;
char message[4096];
- char namebuf[1024+32];
if (isc_log_wouldlog(dns_lctx, level) == ISC_FALSE)
return;
- zone_tostr(zone, namebuf, sizeof(namebuf));
-
va_start(ap, fmt);
vsnprintf(message, sizeof(message), fmt, ap);
va_end(ap);
isc_log_write(dns_lctx, DNS_LOGCATEGORY_GENERAL, DNS_LOGMODULE_ZONE,
- level, "zone %s: %s", namebuf, message);
+ level, "zone %s: %s", zone->strnamerd, message);
}
static void
@@ -6076,19 +9224,16 @@ zone_debuglog(dns_zone_t *zone, const char *me, int debuglevel,
{
va_list ap;
char message[4096];
- char namebuf[1024+32];
int level = ISC_LOG_DEBUG(debuglevel);
if (isc_log_wouldlog(dns_lctx, level) == ISC_FALSE)
return;
- zone_tostr(zone, namebuf, sizeof(namebuf));
-
va_start(ap, fmt);
vsnprintf(message, sizeof(message), fmt, ap);
va_end(ap);
isc_log_write(dns_lctx, DNS_LOGCATEGORY_GENERAL, DNS_LOGMODULE_ZONE,
- level, "%s: zone %s: %s", me, namebuf, message);
+ level, "%s: zone %s: %s", me, zone->strnamerd, message);
}
static int
@@ -6313,12 +9458,16 @@ zone_replacedb(dns_zone_t *zone, dns_db_t *db, isc_boolean_t dump) {
return (result);
}
+ result = check_nsec3param(zone, db);
+ if (result != ISC_R_SUCCESS)
+ return (result);
+
ver = NULL;
dns_db_currentversion(db, &ver);
/*
* The initial version of a slave zone is always dumped;
- * subsequent versions may be journalled instead if this
+ * subsequent versions may be journaled instead if this
* is enabled in the configuration.
*/
if (zone->db != NULL && zone->journal != NULL &&
@@ -6401,7 +9550,7 @@ zone_replacedb(dns_zone_t *zone, dns_db_t *db, isc_boolean_t dump) {
* The in-memory database just changed, and
* because 'dump' is set, it didn't change by
* being loaded from disk. Also, we have not
- * journalled diffs for this change.
+ * journaled diffs for this change.
* Therefore, the on-disk journal is missing
* the deltas for this change. Since it can
* no longer be used to bring the zone
@@ -6411,7 +9560,17 @@ zone_replacedb(dns_zone_t *zone, dns_db_t *db, isc_boolean_t dump) {
isc_log_write(dns_lctx, DNS_LOGCATEGORY_GENERAL,
DNS_LOGMODULE_ZONE, ISC_LOG_DEBUG(3),
"removing journal file");
- (void)remove(zone->journal);
+ if (remove(zone->journal) < 0 && errno != ENOENT) {
+ char strbuf[ISC_STRERRORSIZE];
+ isc__strerror(errno, strbuf, sizeof(strbuf));
+ isc_log_write(dns_lctx,
+ DNS_LOGCATEGORY_GENERAL,
+ DNS_LOGMODULE_ZONE,
+ ISC_LOG_WARNING,
+ "unable to remove journal "
+ "'%s': '%s'",
+ zone->journal, strbuf);
+ }
}
}
@@ -6568,7 +9727,7 @@ zone_xfrdone(dns_zone_t *zone, isc_result_t result) {
}
/*
- * This is not neccessary if we just performed a AXFR
+ * This is not necessary if we just performed a AXFR
* however it is necessary for an IXFR / UPTODATE and
* won't hurt with an AXFR.
*/
@@ -6592,6 +9751,7 @@ zone_xfrdone(dns_zone_t *zone, isc_result_t result) {
dns_result_totext(result));
}
+ inc_stats(zone, dns_zonestatscounter_xfrsuccess);
break;
case DNS_R_BADIXFR:
@@ -6626,6 +9786,7 @@ zone_xfrdone(dns_zone_t *zone, isc_result_t result) {
DNS_ZONE_SETFLAG(zone, DNS_ZONEFLG_REFRESH);
again = ISC_TRUE;
}
+ inc_stats(zone, dns_zonestatscounter_xfrfail);
break;
}
zone_settimer(zone, &now);
@@ -6760,6 +9921,20 @@ dns_zone_getsigvalidityinterval(dns_zone_t *zone) {
return (zone->sigvalidityinterval);
}
+void
+dns_zone_setsigresigninginterval(dns_zone_t *zone, isc_uint32_t interval) {
+ REQUIRE(DNS_ZONE_VALID(zone));
+
+ zone->sigresigninginterval = interval;
+}
+
+isc_uint32_t
+dns_zone_getsigresigninginterval(dns_zone_t *zone) {
+ REQUIRE(DNS_ZONE_VALID(zone));
+
+ return (zone->sigresigninginterval);
+}
+
static void
queue_xfrin(dns_zone_t *zone) {
const char me[] = "queue_xfrin";
@@ -6798,12 +9973,14 @@ static void
got_transfer_quota(isc_task_t *task, isc_event_t *event) {
isc_result_t result;
dns_peer_t *peer = NULL;
- char mastertext[256];
+ char master[ISC_SOCKADDR_FORMATSIZE];
+ char source[ISC_SOCKADDR_FORMATSIZE];
dns_rdatatype_t xfrtype;
dns_zone_t *zone = event->ev_arg;
isc_netaddr_t masterip;
isc_sockaddr_t sourceaddr;
isc_sockaddr_t masteraddr;
+ isc_time_t now;
UNUSED(task);
@@ -6814,34 +9991,44 @@ got_transfer_quota(isc_task_t *task, isc_event_t *event) {
goto cleanup;
}
- isc_sockaddr_format(&zone->masteraddr, mastertext, sizeof(mastertext));
+ TIME_NOW(&now);
+
+ isc_sockaddr_format(&zone->masteraddr, master, sizeof(master));
+ if (dns_zonemgr_unreachable(zone->zmgr, &zone->masteraddr,
+ &zone->sourceaddr, &now)) {
+ isc_sockaddr_format(&zone->sourceaddr, source, sizeof(source));
+ dns_zone_log(zone, ISC_LOG_INFO,
+ "got_transfer_quota: skipping zone transfer as "
+ "master %s (source %s) is unreachable (cached)",
+ master, source);
+ result = ISC_R_CANCELED;
+ goto cleanup;
+ }
isc_netaddr_fromsockaddr(&masterip, &zone->masteraddr);
- (void)dns_peerlist_peerbyaddr(zone->view->peers,
- &masterip, &peer);
+ (void)dns_peerlist_peerbyaddr(zone->view->peers, &masterip, &peer);
/*
* Decide whether we should request IXFR or AXFR.
*/
if (zone->db == NULL) {
dns_zone_log(zone, ISC_LOG_DEBUG(1),
- "no database exists yet, "
- "requesting AXFR of "
- "initial version from %s", mastertext);
+ "no database exists yet, requesting AXFR of "
+ "initial version from %s", master);
xfrtype = dns_rdatatype_axfr;
} else if (DNS_ZONE_OPTION(zone, DNS_ZONEOPT_IXFRFROMDIFFS)) {
dns_zone_log(zone, ISC_LOG_DEBUG(1), "ixfr-from-differences "
- "set, requesting AXFR from %s", mastertext);
+ "set, requesting AXFR from %s", master);
xfrtype = dns_rdatatype_axfr;
} else if (DNS_ZONE_FLAG(zone, DNS_ZONEFLG_FORCEXFER)) {
dns_zone_log(zone, ISC_LOG_DEBUG(1),
"forced reload, requesting AXFR of "
- "initial version from %s", mastertext);
+ "initial version from %s", master);
xfrtype = dns_rdatatype_axfr;
} else if (DNS_ZONE_FLAG(zone, DNS_ZONEFLAG_NOIXFR)) {
dns_zone_log(zone, ISC_LOG_DEBUG(1),
"retrying with AXFR from %s due to "
- "previous IXFR failure", mastertext);
+ "previous IXFR failure", master);
xfrtype = dns_rdatatype_axfr;
LOCK_ZONE(zone);
DNS_ZONE_CLRFLAG(zone, DNS_ZONEFLAG_NOIXFR);
@@ -6857,17 +10044,15 @@ got_transfer_quota(isc_task_t *task, isc_event_t *event) {
}
if (use_ixfr == ISC_FALSE) {
dns_zone_log(zone, ISC_LOG_DEBUG(1),
- "IXFR disabled, "
- "requesting AXFR from %s",
- mastertext);
+ "IXFR disabled, requesting AXFR from %s",
+ master);
if (DNS_ZONE_FLAG(zone, DNS_ZONEFLG_SOABEFOREAXFR))
xfrtype = dns_rdatatype_soa;
else
xfrtype = dns_rdatatype_axfr;
} else {
dns_zone_log(zone, ISC_LOG_DEBUG(1),
- "requesting IXFR from %s",
- mastertext);
+ "requesting IXFR from %s", master);
xfrtype = dns_rdatatype_ixfr;
}
}
@@ -6892,8 +10077,7 @@ got_transfer_quota(isc_task_t *task, isc_event_t *event) {
if (result != ISC_R_SUCCESS && result != ISC_R_NOTFOUND) {
dns_zone_log(zone, ISC_LOG_ERROR,
- "could not get TSIG key "
- "for zone transfer: %s",
+ "could not get TSIG key for zone transfer: %s",
isc_result_totext(result));
}
@@ -6906,6 +10090,21 @@ got_transfer_quota(isc_task_t *task, isc_event_t *event) {
zone->tsigkey, zone->mctx,
zone->zmgr->timermgr, zone->zmgr->socketmgr,
zone->task, zone_xfrdone, &zone->xfr);
+ if (result == ISC_R_SUCCESS) {
+ LOCK_ZONE(zone);
+ if (xfrtype == dns_rdatatype_axfr) {
+ if (isc_sockaddr_pf(&masteraddr) == PF_INET)
+ inc_stats(zone, dns_zonestatscounter_axfrreqv4);
+ else
+ inc_stats(zone, dns_zonestatscounter_axfrreqv6);
+ } else if (xfrtype == dns_rdatatype_ixfr) {
+ if (isc_sockaddr_pf(&masteraddr) == PF_INET)
+ inc_stats(zone, dns_zonestatscounter_ixfrreqv4);
+ else
+ inc_stats(zone, dns_zonestatscounter_ixfrreqv6);
+ }
+ UNLOCK_ZONE(zone);
+ }
cleanup:
/*
* Any failure in this function is handled like a failed
@@ -7175,6 +10374,7 @@ dns_zonemgr_create(isc_mem_t *mctx, isc_taskmgr_t *taskmgr,
ISC_LIST_INIT(zmgr->zones);
ISC_LIST_INIT(zmgr->waiting_for_xfrin);
ISC_LIST_INIT(zmgr->xfrin_in_progress);
+ memset(zmgr->unreachable, 0, sizeof(zmgr->unreachable));
result = isc_rwlock_init(&zmgr->rwlock, 0, 0);
if (result != ISC_R_SUCCESS)
goto free_mem;
@@ -7264,8 +10464,10 @@ dns_zonemgr_managezone(dns_zonemgr_t *zmgr, dns_zone_t *zone) {
NULL, NULL,
zone->task, zone_timer, zone,
&zone->timer);
+
if (result != ISC_R_SUCCESS)
goto cleanup_task;
+
/*
* The timer "holds" a iref.
*/
@@ -7735,7 +10937,7 @@ zone_saveunique(dns_zone_t *zone, const char *path, const char *templat) {
}
#if 0
-/* Hook for ondestroy notifcation from a database. */
+/* Hook for ondestroy notification from a database. */
static void
dns_zonemgr_dbdestroyed(isc_task_t *task, isc_event_t *event) {
@@ -7791,6 +10993,87 @@ dns_zonemgr_getserialqueryrate(dns_zonemgr_t *zmgr) {
return (zmgr->serialqueryrate);
}
+static isc_boolean_t
+dns_zonemgr_unreachable(dns_zonemgr_t *zmgr, isc_sockaddr_t *remote,
+ isc_sockaddr_t *local, isc_time_t *now)
+{
+ unsigned int i;
+ isc_rwlocktype_t locktype;
+ isc_result_t result;
+ isc_uint32_t seconds = isc_time_seconds(now);
+
+ REQUIRE(DNS_ZONEMGR_VALID(zmgr));
+
+ locktype = isc_rwlocktype_read;
+ RWLOCK(&zmgr->rwlock, locktype);
+ for (i = 0; i < UNREACH_CHACHE_SIZE; i++) {
+ if (zmgr->unreachable[i].expire >= seconds &&
+ isc_sockaddr_equal(&zmgr->unreachable[i].remote, remote) &&
+ isc_sockaddr_equal(&zmgr->unreachable[i].local, local)) {
+ result = isc_rwlock_tryupgrade(&zmgr->rwlock);
+ if (result == ISC_R_SUCCESS) {
+ locktype = isc_rwlocktype_write;
+ zmgr->unreachable[i].last = seconds;
+ }
+ break;
+ }
+ }
+ RWUNLOCK(&zmgr->rwlock, locktype);
+ return (ISC_TF(i < UNREACH_CHACHE_SIZE));
+}
+
+void
+dns_zonemgr_unreachableadd(dns_zonemgr_t *zmgr, isc_sockaddr_t *remote,
+ isc_sockaddr_t *local, isc_time_t *now)
+{
+ isc_uint32_t seconds = isc_time_seconds(now);
+ isc_uint32_t last = seconds;
+ unsigned int i, slot = UNREACH_CHACHE_SIZE, oldest = 0;
+
+ REQUIRE(DNS_ZONEMGR_VALID(zmgr));
+
+ RWLOCK(&zmgr->rwlock, isc_rwlocktype_write);
+ for (i = 0; i < UNREACH_CHACHE_SIZE; i++) {
+ /* Existing entry? */
+ if (isc_sockaddr_equal(&zmgr->unreachable[i].remote, remote) &&
+ isc_sockaddr_equal(&zmgr->unreachable[i].local, local))
+ break;
+ /* Empty slot? */
+ if (zmgr->unreachable[i].expire < seconds)
+ slot = i;
+ /* Least recently used slot? */
+ if (zmgr->unreachable[i].last < last) {
+ last = zmgr->unreachable[i].last;
+ oldest = i;
+ }
+ }
+ if (i < UNREACH_CHACHE_SIZE) {
+ /*
+ * Found a existing entry. Update the expire timer and
+ * last usage timestamps.
+ */
+ zmgr->unreachable[i].expire = seconds + UNREACH_HOLD_TIME;
+ zmgr->unreachable[i].last = seconds;
+ } else if (slot != UNREACH_CHACHE_SIZE) {
+ /*
+ * Found a empty slot. Add a new entry to the cache.
+ */
+ zmgr->unreachable[slot].expire = seconds + UNREACH_HOLD_TIME;
+ zmgr->unreachable[slot].last = seconds;
+ zmgr->unreachable[slot].remote = *remote;
+ zmgr->unreachable[slot].local = *local;
+ } else {
+ /*
+ * Replace the least recently used entry in the cache.
+ */
+ zmgr->unreachable[oldest].expire = seconds + UNREACH_HOLD_TIME;
+ zmgr->unreachable[oldest].last = seconds;
+ zmgr->unreachable[oldest].remote = *remote;
+ zmgr->unreachable[oldest].local = *local;
+ }
+ RWUNLOCK(&zmgr->rwlock, isc_rwlocktype_write);
+}
+
void
dns_zone_forcereload(dns_zone_t *zone) {
REQUIRE(DNS_ZONE_VALID(zone));
@@ -7813,26 +11096,66 @@ dns_zone_isforced(dns_zone_t *zone) {
isc_result_t
dns_zone_setstatistics(dns_zone_t *zone, isc_boolean_t on) {
- isc_result_t result = ISC_R_SUCCESS;
+ /*
+ * This function is obsoleted.
+ */
+ UNUSED(zone);
+ UNUSED(on);
+ return (ISC_R_NOTIMPLEMENTED);
+}
+
+isc_uint64_t *
+dns_zone_getstatscounters(dns_zone_t *zone) {
+ /*
+ * This function is obsoleted.
+ */
+ UNUSED(zone);
+ return (NULL);
+}
+
+void
+dns_zone_setstats(dns_zone_t *zone, isc_stats_t *stats) {
+ REQUIRE(DNS_ZONE_VALID(zone));
+ REQUIRE(zone->stats == NULL);
LOCK_ZONE(zone);
- if (on) {
- if (zone->counters != NULL)
- goto done;
- result = dns_stats_alloccounters(zone->mctx, &zone->counters);
- } else {
- if (zone->counters == NULL)
- goto done;
- dns_stats_freecounters(zone->mctx, &zone->counters);
+ zone->stats = NULL;
+ isc_stats_attach(stats, &zone->stats);
+ UNLOCK_ZONE(zone);
+}
+
+void
+dns_zone_setrequeststats(dns_zone_t *zone, isc_stats_t *stats) {
+ REQUIRE(DNS_ZONE_VALID(zone));
+
+ LOCK_ZONE(zone);
+ if (zone->requeststats_on && stats == NULL)
+ zone->requeststats_on = ISC_FALSE;
+ else if (!zone->requeststats_on && stats != NULL) {
+ if (zone->requeststats == NULL) {
+ isc_stats_attach(stats, &zone->requeststats);
+ zone->requeststats_on = ISC_TRUE;
+ }
}
- done:
UNLOCK_ZONE(zone);
- return (result);
+
+ return;
}
-isc_uint64_t *
-dns_zone_getstatscounters(dns_zone_t *zone) {
- return (zone->counters);
+isc_stats_t *
+dns_zone_getrequeststats(dns_zone_t *zone) {
+ /*
+ * We don't lock zone for efficiency reason. This is not catastrophic
+ * because requeststats must always be valid when requeststats_on is
+ * true.
+ * Some counters may be incremented while requeststats_on is becoming
+ * false, or some cannot be incremented just after the statistics are
+ * installed, but it shouldn't matter much in practice.
+ */
+ if (zone->requeststats_on)
+ return (zone->requeststats);
+ else
+ return (NULL);
}
void
@@ -8043,3 +11366,152 @@ dns_zone_getnotifydelay(dns_zone_t *zone) {
return (zone->notifydelay);
}
+
+isc_result_t
+dns_zone_signwithkey(dns_zone_t *zone, dns_secalg_t algorithm,
+ isc_uint16_t keyid, isc_boolean_t delete)
+{
+ isc_result_t result;
+ REQUIRE(DNS_ZONE_VALID(zone));
+
+ dns_zone_log(zone, ISC_LOG_NOTICE,
+ "dns_zone_signwithkey(algorithm=%u, keyid=%u)",
+ algorithm, keyid);
+ LOCK_ZONE(zone);
+ result = zone_signwithkey(zone, algorithm, keyid, delete);
+ UNLOCK_ZONE(zone);
+
+ return (result);
+}
+
+static const char *hex = "0123456789ABCDEF";
+
+isc_result_t
+dns_zone_addnsec3chain(dns_zone_t *zone, dns_rdata_nsec3param_t *nsec3param) {
+ isc_result_t result;
+ char salt[255*2+1];
+ unsigned int i, j;
+
+ REQUIRE(DNS_ZONE_VALID(zone));
+
+ if (nsec3param->salt_length != 0) {
+ INSIST((nsec3param->salt_length * 2U) < sizeof(salt));
+ for (i = 0, j = 0; i < nsec3param->salt_length; i++) {
+ salt[j++] = hex[(nsec3param->salt[i] >> 4) & 0xf];
+ salt[j++] = hex[nsec3param->salt[i] & 0xf];
+ }
+ salt[j] = '\0';
+ } else
+ strcpy(salt, "-");
+ dns_zone_log(zone, ISC_LOG_NOTICE,
+ "dns_zone_addnsec3chain(hash=%u, iterations=%u, salt=%s)",
+ nsec3param->hash, nsec3param->iterations,
+ salt);
+ LOCK_ZONE(zone);
+ result = zone_addnsec3chain(zone, nsec3param);
+ UNLOCK_ZONE(zone);
+
+ return (result);
+}
+
+void
+dns_zone_setnodes(dns_zone_t *zone, isc_uint32_t nodes) {
+ REQUIRE(DNS_ZONE_VALID(zone));
+
+ if (nodes == 0)
+ nodes = 1;
+ zone->nodes = nodes;
+}
+
+void
+dns_zone_setsignatures(dns_zone_t *zone, isc_uint32_t signatures) {
+ REQUIRE(DNS_ZONE_VALID(zone));
+
+ /*
+ * We treat signatures as a signed value so explicitly
+ * limit its range here.
+ */
+ if (signatures > ISC_INT32_MAX)
+ signatures = ISC_INT32_MAX;
+ else if (signatures == 0)
+ signatures = 1;
+ zone->signatures = signatures;
+}
+
+void
+dns_zone_setprivatetype(dns_zone_t *zone, dns_rdatatype_t type) {
+ REQUIRE(DNS_ZONE_VALID(zone));
+ zone->privatetype = type;
+}
+
+dns_rdatatype_t
+dns_zone_getprivatetype(dns_zone_t *zone) {
+ REQUIRE(DNS_ZONE_VALID(zone));
+ return (zone->privatetype);
+}
+
+static isc_result_t
+zone_signwithkey(dns_zone_t *zone, dns_secalg_t algorithm, isc_uint16_t keyid,
+ isc_boolean_t delete)
+{
+ dns_signing_t *signing;
+ dns_signing_t *current;
+ isc_result_t result = ISC_R_SUCCESS;
+ isc_time_t now;
+
+ signing = isc_mem_get(zone->mctx, sizeof *signing);
+ if (signing == NULL)
+ return (ISC_R_NOMEMORY);
+
+ signing->magic = 0;
+ signing->db = NULL;
+ signing->dbiterator = NULL;
+ signing->algorithm = algorithm;
+ signing->keyid = keyid;
+ signing->delete = delete;
+ signing->done = ISC_FALSE;
+
+ TIME_NOW(&now);
+
+ for (current = ISC_LIST_HEAD(zone->signing);
+ current != NULL;
+ current = ISC_LIST_NEXT(current, link)) {
+ if (current->db == zone->db &&
+ current->algorithm == signing->algorithm &&
+ current->keyid == signing->keyid) {
+ if (current->delete != signing->delete)
+ current->done = ISC_TRUE;
+ else
+ goto cleanup;
+ }
+ }
+
+ if (zone->db != NULL) {
+ dns_db_attach(zone->db, &signing->db);
+ result = dns_db_createiterator(signing->db, 0,
+ &signing->dbiterator);
+
+ if (result == ISC_R_SUCCESS)
+ result = dns_dbiterator_first(signing->dbiterator);
+ if (result == ISC_R_SUCCESS) {
+ dns_dbiterator_pause(signing->dbiterator);
+ ISC_LIST_INITANDAPPEND(zone->signing, signing, link);
+ signing = NULL;
+ if (isc_time_isepoch(&zone->signingtime)) {
+ zone->signingtime = now;
+ if (zone->task != NULL)
+ zone_settimer(zone, &now);
+ }
+ }
+ } else
+ result = ISC_R_NOTFOUND;
+
+ cleanup:
+ if (signing != NULL) {
+ dns_db_detach(&signing->db);
+ if (signing->dbiterator != NULL)
+ dns_dbiterator_destroy(&signing->dbiterator);
+ isc_mem_put(zone->mctx, signing, sizeof *signing);
+ }
+ return (result);
+}
OpenPOWER on IntegriCloud