/* * Copyright (C) 2004 Internet Systems Consortium, Inc. ("ISC") * Copyright (C) 1999-2003 Internet Software Consortium. * * Permission to use, copy, modify, and distribute this software for any * purpose with or without fee is hereby granted, provided that the above * copyright notice and this permission notice appear in all copies. * * THE SOFTWARE IS PROVIDED "AS IS" AND ISC DISCLAIMS ALL WARRANTIES WITH * REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY * AND FITNESS. IN NO EVENT SHALL ISC BE LIABLE FOR ANY SPECIAL, DIRECT, * INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM * LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE * OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR * PERFORMANCE OF THIS SOFTWARE. */ /* $ISC: server.c,v 1.339.2.15.2.59 2004/11/10 22:13:56 marka Exp $ */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * Check an operation for failure. Assumes that the function * using it has a 'result' variable and a 'cleanup' label. */ #define CHECK(op) \ do { result = (op); \ if (result != ISC_R_SUCCESS) goto cleanup; \ } while (0) #define CHECKM(op, msg) \ do { result = (op); \ if (result != ISC_R_SUCCESS) { \ isc_log_write(ns_g_lctx, \ NS_LOGCATEGORY_GENERAL, \ NS_LOGMODULE_SERVER, \ ISC_LOG_ERROR, \ "%s: %s", msg, \ isc_result_totext(result)); \ goto cleanup; \ } \ } while (0) \ #define CHECKMF(op, msg, file) \ do { result = (op); \ if (result != ISC_R_SUCCESS) { \ isc_log_write(ns_g_lctx, \ NS_LOGCATEGORY_GENERAL, \ NS_LOGMODULE_SERVER, \ ISC_LOG_ERROR, \ "%s '%s': %s", msg, file, \ isc_result_totext(result)); \ goto cleanup; \ } \ } while (0) \ #define CHECKFATAL(op, msg) \ do { result = (op); \ if (result != ISC_R_SUCCESS) \ fatal(msg, result); \ } while (0) \ struct ns_dispatch { isc_sockaddr_t addr; unsigned int dispatchgen; dns_dispatch_t *dispatch; ISC_LINK(struct ns_dispatch) link; }; struct dumpcontext { isc_mem_t *mctx; isc_boolean_t dumpcache; isc_boolean_t dumpzones; FILE *fp; ISC_LIST(struct viewlistentry) viewlist; struct viewlistentry *view; struct zonelistentry *zone; dns_dumpctx_t *mdctx; dns_db_t *db; dns_db_t *cache; isc_task_t *task; dns_dbversion_t *version; }; struct viewlistentry { dns_view_t *view; ISC_LINK(struct viewlistentry) link; ISC_LIST(struct zonelistentry) zonelist; }; struct zonelistentry { dns_zone_t *zone; ISC_LINK(struct zonelistentry) link; }; static void fatal(const char *msg, isc_result_t result); static void ns_server_reload(isc_task_t *task, isc_event_t *event); static isc_result_t ns_listenelt_fromconfig(cfg_obj_t *listener, cfg_obj_t *config, ns_aclconfctx_t *actx, isc_mem_t *mctx, ns_listenelt_t **target); static isc_result_t ns_listenlist_fromconfig(cfg_obj_t *listenlist, cfg_obj_t *config, ns_aclconfctx_t *actx, isc_mem_t *mctx, ns_listenlist_t **target); static isc_result_t configure_forward(cfg_obj_t *config, dns_view_t *view, dns_name_t *origin, cfg_obj_t *forwarders, cfg_obj_t *forwardtype); static isc_result_t configure_alternates(cfg_obj_t *config, dns_view_t *view, cfg_obj_t *alternates); static isc_result_t configure_zone(cfg_obj_t *config, cfg_obj_t *zconfig, cfg_obj_t *vconfig, isc_mem_t *mctx, dns_view_t *view, ns_aclconfctx_t *aclconf); static void end_reserved_dispatches(ns_server_t *server, isc_boolean_t all); /* * Configure a single view ACL at '*aclp'. Get its configuration by * calling 'getvcacl' (for per-view configuration) and maybe 'getscacl' * (for a global default). */ static isc_result_t configure_view_acl(cfg_obj_t *vconfig, cfg_obj_t *config, const char *aclname, ns_aclconfctx_t *actx, isc_mem_t *mctx, dns_acl_t **aclp) { isc_result_t result; cfg_obj_t *maps[3]; cfg_obj_t *aclobj = NULL; int i = 0; if (*aclp != NULL) dns_acl_detach(aclp); if (vconfig != NULL) maps[i++] = cfg_tuple_get(vconfig, "options"); if (config != NULL) { cfg_obj_t *options = NULL; (void)cfg_map_get(config, "options", &options); if (options != NULL) maps[i++] = options; } maps[i] = NULL; result = ns_config_get(maps, aclname, &aclobj); if (aclobj == NULL) /* * No value available. *aclp == NULL. */ return (ISC_R_SUCCESS); result = ns_acl_fromconfig(aclobj, config, actx, mctx, aclp); return (result); } static isc_result_t configure_view_dnsseckey(cfg_obj_t *vconfig, cfg_obj_t *key, dns_keytable_t *keytable, isc_mem_t *mctx) { dns_rdataclass_t viewclass; dns_rdata_dnskey_t keystruct; isc_uint32_t flags, proto, alg; char *keystr, *keynamestr; unsigned char keydata[4096]; isc_buffer_t keydatabuf; unsigned char rrdata[4096]; isc_buffer_t rrdatabuf; isc_region_t r; dns_fixedname_t fkeyname; dns_name_t *keyname; isc_buffer_t namebuf; isc_result_t result; dst_key_t *dstkey = NULL; flags = cfg_obj_asuint32(cfg_tuple_get(key, "flags")); proto = cfg_obj_asuint32(cfg_tuple_get(key, "protocol")); alg = cfg_obj_asuint32(cfg_tuple_get(key, "algorithm")); keyname = dns_fixedname_name(&fkeyname); keynamestr = cfg_obj_asstring(cfg_tuple_get(key, "name")); if (vconfig == NULL) viewclass = dns_rdataclass_in; else { cfg_obj_t *classobj = cfg_tuple_get(vconfig, "class"); CHECK(ns_config_getclass(classobj, dns_rdataclass_in, &viewclass)); } keystruct.common.rdclass = viewclass; keystruct.common.rdtype = dns_rdatatype_dnskey; /* * The key data in keystruct is not dynamically allocated. */ keystruct.mctx = NULL; ISC_LINK_INIT(&keystruct.common, link); if (flags > 0xffff) CHECKM(ISC_R_RANGE, "key flags"); if (proto > 0xff) CHECKM(ISC_R_RANGE, "key protocol"); if (alg > 0xff) CHECKM(ISC_R_RANGE, "key algorithm"); keystruct.flags = (isc_uint16_t)flags; keystruct.protocol = (isc_uint8_t)proto; keystruct.algorithm = (isc_uint8_t)alg; isc_buffer_init(&keydatabuf, keydata, sizeof(keydata)); isc_buffer_init(&rrdatabuf, rrdata, sizeof(rrdata)); keystr = cfg_obj_asstring(cfg_tuple_get(key, "key")); CHECK(isc_base64_decodestring(keystr, &keydatabuf)); isc_buffer_usedregion(&keydatabuf, &r); keystruct.datalen = r.length; keystruct.data = r.base; CHECK(dns_rdata_fromstruct(NULL, keystruct.common.rdclass, keystruct.common.rdtype, &keystruct, &rrdatabuf)); dns_fixedname_init(&fkeyname); isc_buffer_init(&namebuf, keynamestr, strlen(keynamestr)); isc_buffer_add(&namebuf, strlen(keynamestr)); CHECK(dns_name_fromtext(keyname, &namebuf, dns_rootname, ISC_FALSE, NULL)); CHECK(dst_key_fromdns(keyname, viewclass, &rrdatabuf, mctx, &dstkey)); CHECK(dns_keytable_add(keytable, &dstkey)); INSIST(dstkey == NULL); return (ISC_R_SUCCESS); cleanup: if (result == DST_R_NOCRYPTO) { cfg_obj_log(key, ns_g_lctx, ISC_LOG_ERROR, "ignoring trusted key for '%s': no crypto support", keynamestr); result = ISC_R_SUCCESS; } else { cfg_obj_log(key, ns_g_lctx, ISC_LOG_ERROR, "configuring trusted key for '%s': %s", keynamestr, isc_result_totext(result)); result = ISC_R_FAILURE; } if (dstkey != NULL) dst_key_free(&dstkey); return (result); } /* * Configure DNSSEC keys for a view. Currently used only for * the security roots. * * The per-view configuration values and the server-global defaults are read * from 'vconfig' and 'config'. The variable to be configured is '*target'. */ static isc_result_t configure_view_dnsseckeys(cfg_obj_t *vconfig, cfg_obj_t *config, isc_mem_t *mctx, dns_keytable_t **target) { isc_result_t result; cfg_obj_t *keys = NULL; cfg_obj_t *voptions = NULL; cfg_listelt_t *element, *element2; cfg_obj_t *keylist; cfg_obj_t *key; dns_keytable_t *keytable = NULL; CHECK(dns_keytable_create(mctx, &keytable)); if (vconfig != NULL) voptions = cfg_tuple_get(vconfig, "options"); keys = NULL; if (voptions != NULL) (void)cfg_map_get(voptions, "trusted-keys", &keys); if (keys == NULL) (void)cfg_map_get(config, "trusted-keys", &keys); for (element = cfg_list_first(keys); element != NULL; element = cfg_list_next(element)) { keylist = cfg_listelt_value(element); for (element2 = cfg_list_first(keylist); element2 != NULL; element2 = cfg_list_next(element2)) { key = cfg_listelt_value(element2); CHECK(configure_view_dnsseckey(vconfig, key, keytable, mctx)); } } dns_keytable_detach(target); *target = keytable; /* Transfer ownership. */ keytable = NULL; result = ISC_R_SUCCESS; cleanup: return (result); } static isc_result_t mustbesecure(cfg_obj_t *mbs, dns_resolver_t *resolver) { cfg_listelt_t *element; cfg_obj_t *obj; const char *str; dns_fixedname_t fixed; dns_name_t *name; isc_boolean_t value; isc_result_t result; isc_buffer_t b; dns_fixedname_init(&fixed); name = dns_fixedname_name(&fixed); for (element = cfg_list_first(mbs); element != NULL; element = cfg_list_next(element)) { obj = cfg_listelt_value(element); str = cfg_obj_asstring(cfg_tuple_get(obj, "name")); isc_buffer_init(&b, str, strlen(str)); isc_buffer_add(&b, strlen(str)); CHECK(dns_name_fromtext(name, &b, dns_rootname, ISC_FALSE, NULL)); value = cfg_obj_asboolean(cfg_tuple_get(obj, "value")); CHECK(dns_resolver_setmustbesecure(resolver, name, value)); } result = ISC_R_SUCCESS; cleanup: return (result); } /* * Get a dispatch appropriate for the resolver of a given view. */ static isc_result_t get_view_querysource_dispatch(cfg_obj_t **maps, int af, dns_dispatch_t **dispatchp) { isc_result_t result; dns_dispatch_t *disp; isc_sockaddr_t sa; unsigned int attrs, attrmask; cfg_obj_t *obj = NULL; /* * Make compiler happy. */ result = ISC_R_FAILURE; switch (af) { case AF_INET: result = ns_config_get(maps, "query-source", &obj); INSIST(result == ISC_R_SUCCESS); break; case AF_INET6: result = ns_config_get(maps, "query-source-v6", &obj); INSIST(result == ISC_R_SUCCESS); break; default: INSIST(0); } sa = *(cfg_obj_assockaddr(obj)); INSIST(isc_sockaddr_pf(&sa) == af); /* * If we don't support this address family, we're done! */ switch (af) { case AF_INET: result = isc_net_probeipv4(); break; case AF_INET6: result = isc_net_probeipv6(); break; default: INSIST(0); } if (result != ISC_R_SUCCESS) return (ISC_R_SUCCESS); /* * Try to find a dispatcher that we can share. */ attrs = 0; attrs |= DNS_DISPATCHATTR_UDP; switch (af) { case AF_INET: attrs |= DNS_DISPATCHATTR_IPV4; break; case AF_INET6: attrs |= DNS_DISPATCHATTR_IPV6; break; } attrmask = 0; attrmask |= DNS_DISPATCHATTR_UDP; attrmask |= DNS_DISPATCHATTR_TCP; attrmask |= DNS_DISPATCHATTR_IPV4; attrmask |= DNS_DISPATCHATTR_IPV6; disp = NULL; result = dns_dispatch_getudp(ns_g_dispatchmgr, ns_g_socketmgr, ns_g_taskmgr, &sa, 4096, 1000, 32768, 16411, 16433, attrs, attrmask, &disp); if (result != ISC_R_SUCCESS) { isc_sockaddr_t any; char buf[ISC_SOCKADDR_FORMATSIZE]; switch (af) { case AF_INET: isc_sockaddr_any(&any); break; case AF_INET6: isc_sockaddr_any6(&any); break; } if (isc_sockaddr_equal(&sa, &any)) return (ISC_R_SUCCESS); isc_sockaddr_format(&sa, buf, sizeof(buf)); isc_log_write(ns_g_lctx, NS_LOGCATEGORY_GENERAL, NS_LOGMODULE_SERVER, ISC_LOG_ERROR, "could not get query source dispatcher (%s)", buf); return (result); } *dispatchp = disp; return (ISC_R_SUCCESS); } static isc_result_t configure_order(dns_order_t *order, cfg_obj_t *ent) { dns_rdataclass_t rdclass; dns_rdatatype_t rdtype; cfg_obj_t *obj; dns_fixedname_t fixed; unsigned int mode = 0; const char *str; isc_buffer_t b; isc_result_t result; isc_boolean_t addroot; result = ns_config_getclass(cfg_tuple_get(ent, "class"), dns_rdataclass_any, &rdclass); if (result != ISC_R_SUCCESS) return (result); result = ns_config_gettype(cfg_tuple_get(ent, "type"), dns_rdatatype_any, &rdtype); if (result != ISC_R_SUCCESS) return (result); obj = cfg_tuple_get(ent, "name"); if (cfg_obj_isstring(obj)) str = cfg_obj_asstring(obj); else str = "*"; addroot = ISC_TF(strcmp(str, "*") == 0); isc_buffer_init(&b, str, strlen(str)); isc_buffer_add(&b, strlen(str)); dns_fixedname_init(&fixed); result = dns_name_fromtext(dns_fixedname_name(&fixed), &b, dns_rootname, ISC_FALSE, NULL); if (result != ISC_R_SUCCESS) return (result); obj = cfg_tuple_get(ent, "ordering"); INSIST(cfg_obj_isstring(obj)); str = cfg_obj_asstring(obj); if (!strcasecmp(str, "fixed")) mode = DNS_RDATASETATTR_FIXEDORDER; else if (!strcasecmp(str, "random")) mode = DNS_RDATASETATTR_RANDOMIZE; else if (!strcasecmp(str, "cyclic")) mode = 0; else INSIST(0); /* * "*" should match everything including the root (BIND 8 compat). * As dns_name_matcheswildcard(".", "*.") returns FALSE add a * explict entry for "." when the name is "*". */ if (addroot) { result = dns_order_add(order, dns_rootname, rdtype, rdclass, mode); if (result != ISC_R_SUCCESS) return (result); } return (dns_order_add(order, dns_fixedname_name(&fixed), rdtype, rdclass, mode)); } static isc_result_t configure_peer(cfg_obj_t *cpeer, isc_mem_t *mctx, dns_peer_t **peerp) { isc_sockaddr_t *sa; isc_netaddr_t na; dns_peer_t *peer; cfg_obj_t *obj; char *str; isc_result_t result; sa = cfg_obj_assockaddr(cfg_map_getname(cpeer)); isc_netaddr_fromsockaddr(&na, sa); peer = NULL; result = dns_peer_new(mctx, &na, &peer); if (result != ISC_R_SUCCESS) return (result); obj = NULL; (void)cfg_map_get(cpeer, "bogus", &obj); if (obj != NULL) CHECK(dns_peer_setbogus(peer, cfg_obj_asboolean(obj))); obj = NULL; (void)cfg_map_get(cpeer, "provide-ixfr", &obj); if (obj != NULL) CHECK(dns_peer_setprovideixfr(peer, cfg_obj_asboolean(obj))); obj = NULL; (void)cfg_map_get(cpeer, "request-ixfr", &obj); if (obj != NULL) CHECK(dns_peer_setrequestixfr(peer, cfg_obj_asboolean(obj))); obj = NULL; (void)cfg_map_get(cpeer, "edns", &obj); if (obj != NULL) CHECK(dns_peer_setsupportedns(peer, cfg_obj_asboolean(obj))); obj = NULL; (void)cfg_map_get(cpeer, "transfers", &obj); if (obj != NULL) CHECK(dns_peer_settransfers(peer, cfg_obj_asuint32(obj))); obj = NULL; (void)cfg_map_get(cpeer, "transfer-format", &obj); if (obj != NULL) { str = cfg_obj_asstring(obj); if (strcasecmp(str, "many-answers") == 0) CHECK(dns_peer_settransferformat(peer, dns_many_answers)); else if (strcasecmp(str, "one-answer") == 0) CHECK(dns_peer_settransferformat(peer, dns_one_answer)); else INSIST(0); } obj = NULL; (void)cfg_map_get(cpeer, "keys", &obj); if (obj != NULL) { result = dns_peer_setkeybycharp(peer, cfg_obj_asstring(obj)); if (result != ISC_R_SUCCESS) goto cleanup; } obj = NULL; if (isc_sockaddr_pf(sa) == AF_INET) (void)cfg_map_get(cpeer, "transfer-source", &obj); else (void)cfg_map_get(cpeer, "transfer-source-v6", &obj); if (obj != NULL) { result = dns_peer_settransfersource(peer, cfg_obj_assockaddr(obj)); if (result != ISC_R_SUCCESS) goto cleanup; } *peerp = peer; return (ISC_R_SUCCESS); cleanup: dns_peer_detach(&peer); return (result); } static isc_result_t disable_algorithms(cfg_obj_t *disabled, dns_resolver_t *resolver) { isc_result_t result; cfg_obj_t *algorithms; cfg_listelt_t *element; const char *str; dns_fixedname_t fixed; dns_name_t *name; isc_buffer_t b; dns_fixedname_init(&fixed); name = dns_fixedname_name(&fixed); str = cfg_obj_asstring(cfg_tuple_get(disabled, "name")); isc_buffer_init(&b, str, strlen(str)); isc_buffer_add(&b, strlen(str)); CHECK(dns_name_fromtext(name, &b, dns_rootname, ISC_FALSE, NULL)); algorithms = cfg_tuple_get(disabled, "algorithms"); for (element = cfg_list_first(algorithms); element != NULL; element = cfg_list_next(element)) { isc_textregion_t r; dns_secalg_t alg; r.base = cfg_obj_asstring(cfg_listelt_value(element)); r.length = strlen(r.base); result = dns_secalg_fromtext(&alg, &r); if (result != ISC_R_SUCCESS) { isc_uint8_t ui; result = isc_parse_uint8(&ui, r.base, 10); alg = ui; } if (result != ISC_R_SUCCESS) { cfg_obj_log(cfg_listelt_value(element), ns_g_lctx, ISC_LOG_ERROR, "invalid algorithm"); CHECK(result); } CHECK(dns_resolver_disable_algorithm(resolver, name, alg)); } cleanup: return (result); } /* * Configure 'view' according to 'vconfig', taking defaults from 'config' * where values are missing in 'vconfig'. * * When configuring the default view, 'vconfig' will be NULL and the * global defaults in 'config' used exclusively. */ static isc_result_t configure_view(dns_view_t *view, cfg_obj_t *config, cfg_obj_t *vconfig, isc_mem_t *mctx, ns_aclconfctx_t *actx, isc_boolean_t need_hints) { cfg_obj_t *maps[4]; cfg_obj_t *cfgmaps[3]; cfg_obj_t *options = NULL; cfg_obj_t *voptions = NULL; cfg_obj_t *forwardtype; cfg_obj_t *forwarders; cfg_obj_t *alternates; cfg_obj_t *zonelist; cfg_obj_t *disabled; cfg_obj_t *obj; cfg_listelt_t *element; in_port_t port; dns_cache_t *cache = NULL; isc_result_t result; isc_uint32_t max_adb_size; isc_uint32_t max_cache_size; isc_uint32_t lame_ttl; dns_tsig_keyring_t *ring; dns_view_t *pview = NULL; /* Production view */ isc_mem_t *cmctx; dns_dispatch_t *dispatch4 = NULL; dns_dispatch_t *dispatch6 = NULL; isc_boolean_t reused_cache = ISC_FALSE; int i; const char *str; dns_order_t *order = NULL; isc_uint32_t udpsize; unsigned int check = 0; REQUIRE(DNS_VIEW_VALID(view)); cmctx = NULL; if (config != NULL) (void)cfg_map_get(config, "options", &options); i = 0; if (vconfig != NULL) { voptions = cfg_tuple_get(vconfig, "options"); maps[i++] = voptions; } if (options != NULL) maps[i++] = options; maps[i++] = ns_g_defaults; maps[i] = NULL; i = 0; if (voptions != NULL) cfgmaps[i++] = voptions; if (config != NULL) cfgmaps[i++] = config; cfgmaps[i] = NULL; /* * Set the view's port number for outgoing queries. */ CHECKM(ns_config_getport(config, &port), "port"); dns_view_setdstport(view, port); /* * Configure the zones. */ zonelist = NULL; if (voptions != NULL) (void)cfg_map_get(voptions, "zone", &zonelist); else (void)cfg_map_get(config, "zone", &zonelist); for (element = cfg_list_first(zonelist); element != NULL; element = cfg_list_next(element)) { cfg_obj_t *zconfig = cfg_listelt_value(element); CHECK(configure_zone(config, zconfig, vconfig, mctx, view, actx)); } /* * Configure the view's cache. Try to reuse an existing * cache if possible, otherwise create a new cache. * Note that the ADB is not preserved in either case. * * XXX Determining when it is safe to reuse a cache is * tricky. When the view's configuration changes, the cached * data may become invalid because it reflects our old * view of the world. As more view attributes become * configurable, we will have to add code here to check * whether they have changed in ways that could * invalidate the cache. */ result = dns_viewlist_find(&ns_g_server->viewlist, view->name, view->rdclass, &pview); if (result != ISC_R_NOTFOUND && result != ISC_R_SUCCESS) goto cleanup; if (pview != NULL) { INSIST(pview->cache != NULL); isc_log_write(ns_g_lctx, NS_LOGCATEGORY_GENERAL, NS_LOGMODULE_SERVER, ISC_LOG_DEBUG(3), "reusing existing cache"); reused_cache = ISC_TRUE; dns_cache_attach(pview->cache, &cache); dns_view_detach(&pview); } else { CHECK(isc_mem_create(0, 0, &cmctx)); CHECK(dns_cache_create(cmctx, ns_g_taskmgr, ns_g_timermgr, view->rdclass, "rbt", 0, NULL, &cache)); } dns_view_setcache(view, cache); /* * cache-file cannot be inherited if views are present, but this * should be caught by the configuration checking stage. */ obj = NULL; result = ns_config_get(maps, "cache-file", &obj); if (result == ISC_R_SUCCESS && strcmp(view->name, "_bind") != 0) { CHECK(dns_cache_setfilename(cache, cfg_obj_asstring(obj))); if (!reused_cache) CHECK(dns_cache_load(cache)); } obj = NULL; result = ns_config_get(maps, "cleaning-interval", &obj); INSIST(result == ISC_R_SUCCESS); dns_cache_setcleaninginterval(cache, cfg_obj_asuint32(obj) * 60); obj = NULL; result = ns_config_get(maps, "max-cache-size", &obj); INSIST(result == ISC_R_SUCCESS); if (cfg_obj_isstring(obj)) { str = cfg_obj_asstring(obj); INSIST(strcasecmp(str, "unlimited") == 0); max_cache_size = ISC_UINT32_MAX; } else { isc_resourcevalue_t value; value = cfg_obj_asuint64(obj); if (value > ISC_UINT32_MAX) { cfg_obj_log(obj, ns_g_lctx, ISC_LOG_ERROR, "'max-cache-size " "%" ISC_PRINT_QUADFORMAT "d' is too large", value); result = ISC_R_RANGE; goto cleanup; } max_cache_size = (isc_uint32_t)value; } dns_cache_setcachesize(cache, max_cache_size); dns_cache_detach(&cache); /* * Check-names. */ obj = NULL; result = ns_checknames_get(maps, "response", &obj); INSIST(result == ISC_R_SUCCESS); str = cfg_obj_asstring(obj); if (strcasecmp(str, "fail") == 0) { check = DNS_RESOLVER_CHECKNAMES | DNS_RESOLVER_CHECKNAMESFAIL; view->checknames = ISC_TRUE; } else if (strcasecmp(str, "warn") == 0) { check = DNS_RESOLVER_CHECKNAMES; view->checknames = ISC_FALSE; } else if (strcasecmp(str, "ignore") == 0) { check = 0; view->checknames = ISC_FALSE; } else INSIST(0); /* * Resolver. * * XXXRTH Hardwired number of tasks. */ CHECK(get_view_querysource_dispatch(maps, AF_INET, &dispatch4)); CHECK(get_view_querysource_dispatch(maps, AF_INET6, &dispatch6)); if (dispatch4 == NULL && dispatch6 == NULL) { UNEXPECTED_ERROR(__FILE__, __LINE__, "unable to obtain neither an IPv4 nor" " an IPv6 dispatch"); result = ISC_R_UNEXPECTED; goto cleanup; } CHECK(dns_view_createresolver(view, ns_g_taskmgr, 31, ns_g_socketmgr, ns_g_timermgr, check, ns_g_dispatchmgr, dispatch4, dispatch6)); /* * Set the ADB cache size to 1/8th of the max-cache-size. */ max_adb_size = 0; if (max_cache_size != 0) { max_adb_size = max_cache_size / 8; if (max_adb_size == 0) max_adb_size = 1; /* Force minimum. */ } dns_adb_setadbsize(view->adb, max_adb_size); /* * Set resolver's lame-ttl. */ obj = NULL; result = ns_config_get(maps, "lame-ttl", &obj); INSIST(result == ISC_R_SUCCESS); lame_ttl = cfg_obj_asuint32(obj); if (lame_ttl > 1800) lame_ttl = 1800; dns_resolver_setlamettl(view->resolver, lame_ttl); /* * Set the resolver's EDNS UDP size. */ obj = NULL; result = ns_config_get(maps, "edns-udp-size", &obj); INSIST(result == ISC_R_SUCCESS); udpsize = cfg_obj_asuint32(obj); if (udpsize < 512) udpsize = 512; if (udpsize > 4096) udpsize = 4096; dns_resolver_setudpsize(view->resolver, (isc_uint16_t)udpsize); /* * Set supported DNSSEC algorithms. */ dns_resolver_reset_algorithms(view->resolver); disabled = NULL; (void)ns_config_get(maps, "disable-algorithms", &disabled); if (disabled != NULL) { for (element = cfg_list_first(disabled); element != NULL; element = cfg_list_next(element)) CHECK(disable_algorithms(cfg_listelt_value(element), view->resolver)); } /* * A global or view "forwarders" option, if present, * creates an entry for "." in the forwarding table. */ forwardtype = NULL; forwarders = NULL; (void)ns_config_get(maps, "forward", &forwardtype); (void)ns_config_get(maps, "forwarders", &forwarders); if (forwarders != NULL) CHECK(configure_forward(config, view, dns_rootname, forwarders, forwardtype)); /* * Dual Stack Servers. */ alternates = NULL; (void)ns_config_get(maps, "dual-stack-servers", &alternates); if (alternates != NULL) CHECK(configure_alternates(config, view, alternates)); /* * We have default hints for class IN if we need them. */ if (view->rdclass == dns_rdataclass_in && view->hints == NULL) dns_view_sethints(view, ns_g_server->in_roothints); /* * If we still have no hints, this is a non-IN view with no * "hints zone" configured. Issue a warning, except if this * is a root server. Root servers never need to consult * their hints, so it's no point requiring users to configure * them. */ if (view->hints == NULL) { dns_zone_t *rootzone = NULL; (void)dns_view_findzone(view, dns_rootname, &rootzone); if (rootzone != NULL) { dns_zone_detach(&rootzone); need_hints = ISC_FALSE; } if (need_hints) isc_log_write(ns_g_lctx, NS_LOGCATEGORY_GENERAL, NS_LOGMODULE_SERVER, ISC_LOG_WARNING, "no root hints for view '%s'", view->name); } /* * Configure the view's TSIG keys. */ ring = NULL; CHECK(ns_tsigkeyring_fromconfig(config, vconfig, view->mctx, &ring)); dns_view_setkeyring(view, ring); /* * Configure the view's peer list. */ { cfg_obj_t *peers = NULL; cfg_listelt_t *element; dns_peerlist_t *newpeers = NULL; (void)ns_config_get(cfgmaps, "server", &peers); CHECK(dns_peerlist_new(mctx, &newpeers)); for (element = cfg_list_first(peers); element != NULL; element = cfg_list_next(element)) { cfg_obj_t *cpeer = cfg_listelt_value(element); dns_peer_t *peer; CHECK(configure_peer(cpeer, mctx, &peer)); dns_peerlist_addpeer(newpeers, peer); dns_peer_detach(&peer); } dns_peerlist_detach(&view->peers); view->peers = newpeers; /* Transfer ownership. */ } /* * Configure the views rrset-order. */ { cfg_obj_t *rrsetorder = NULL; cfg_listelt_t *element; (void)ns_config_get(maps, "rrset-order", &rrsetorder); CHECK(dns_order_create(mctx, &order)); for (element = cfg_list_first(rrsetorder); element != NULL; element = cfg_list_next(element)) { cfg_obj_t *ent = cfg_listelt_value(element); CHECK(configure_order(order, ent)); } if (view->order != NULL) dns_order_detach(&view->order); dns_order_attach(order, &view->order); dns_order_detach(&order); } /* * Copy the aclenv object. */ dns_aclenv_copy(&view->aclenv, &ns_g_server->aclenv); /* * Configure the "match-clients" and "match-destinations" ACL. */ CHECK(configure_view_acl(vconfig, config, "match-clients", actx, ns_g_mctx, &view->matchclients)); CHECK(configure_view_acl(vconfig, config, "match-destinations", actx, ns_g_mctx, &view->matchdestinations)); /* * Configure the "match-recursive-only" option. */ obj = NULL; (void) ns_config_get(maps, "match-recursive-only", &obj); if (obj != NULL && cfg_obj_asboolean(obj)) view->matchrecursiveonly = ISC_TRUE; else view->matchrecursiveonly = ISC_FALSE; /* * Configure other configurable data. */ obj = NULL; result = ns_config_get(maps, "recursion", &obj); INSIST(result == ISC_R_SUCCESS); view->recursion = cfg_obj_asboolean(obj); obj = NULL; result = ns_config_get(maps, "auth-nxdomain", &obj); INSIST(result == ISC_R_SUCCESS); view->auth_nxdomain = cfg_obj_asboolean(obj); obj = NULL; result = ns_config_get(maps, "minimal-responses", &obj); INSIST(result == ISC_R_SUCCESS); view->minimalresponses = cfg_obj_asboolean(obj); obj = NULL; result = ns_config_get(maps, "transfer-format", &obj); INSIST(result == ISC_R_SUCCESS); str = cfg_obj_asstring(obj); if (strcasecmp(str, "many-answers") == 0) view->transfer_format = dns_many_answers; else if (strcasecmp(str, "one-answer") == 0) view->transfer_format = dns_one_answer; else INSIST(0); /* * Set sources where additional data and CNAME/DNAME * targets for authoritative answers may be found. */ obj = NULL; result = ns_config_get(maps, "additional-from-auth", &obj); INSIST(result == ISC_R_SUCCESS); view->additionalfromauth = cfg_obj_asboolean(obj); if (view->recursion && ! view->additionalfromauth) { cfg_obj_log(obj, ns_g_lctx, ISC_LOG_WARNING, "'additional-from-auth no' is only supported " "with 'recursion no'"); view->additionalfromauth = ISC_TRUE; } obj = NULL; result = ns_config_get(maps, "additional-from-cache", &obj); INSIST(result == ISC_R_SUCCESS); view->additionalfromcache = cfg_obj_asboolean(obj); if (view->recursion && ! view->additionalfromcache) { cfg_obj_log(obj, ns_g_lctx, ISC_LOG_WARNING, "'additional-from-cache no' is only supported " "with 'recursion no'"); view->additionalfromcache = ISC_TRUE; } CHECK(configure_view_acl(vconfig, config, "allow-query", actx, ns_g_mctx, &view->queryacl)); if (strcmp(view->name, "_bind") != 0) CHECK(configure_view_acl(vconfig, config, "allow-recursion", actx, ns_g_mctx, &view->recursionacl)); /* * Warning if both "recursion no;" and allow-recursion are active * except for "allow-recursion { none; };". */ if (!view->recursion && view->recursionacl != NULL && (view->recursionacl->length != 1 || view->recursionacl->elements[0].type != dns_aclelementtype_any || view->recursionacl->elements[0].negative != ISC_TRUE)) { const char *forview = " for view "; const char *viewname = view->name; if (!strcmp(view->name, "_bind") || !strcmp(view->name, "_default")) { forview = ""; viewname = ""; } isc_log_write(ns_g_lctx, NS_LOGCATEGORY_GENERAL, NS_LOGMODULE_SERVER, ISC_LOG_WARNING, "both \"recursion no;\" and \"allow-recursion\" " "active%s%s", forview, viewname); } CHECK(configure_view_acl(vconfig, config, "sortlist", actx, ns_g_mctx, &view->sortlist)); obj = NULL; result = ns_config_get(maps, "request-ixfr", &obj); INSIST(result == ISC_R_SUCCESS); view->requestixfr = cfg_obj_asboolean(obj); obj = NULL; result = ns_config_get(maps, "provide-ixfr", &obj); INSIST(result == ISC_R_SUCCESS); view->provideixfr = cfg_obj_asboolean(obj); obj = NULL; result = ns_config_get(maps, "dnssec-enable", &obj); INSIST(result == ISC_R_SUCCESS); view->enablednssec = cfg_obj_asboolean(obj); obj = NULL; result = ns_config_get(maps, "dnssec-lookaside", &obj); if (result == ISC_R_SUCCESS) { for (element = cfg_list_first(obj); element != NULL; element = cfg_list_next(element)) { const char *str; isc_buffer_t b; dns_name_t *dlv; obj = cfg_listelt_value(element); #if 0 dns_fixedname_t fixed; dns_name_t *name; /* * When we support multiple dnssec-lookaside * entries this is how to find the domain to be * checked. XXXMPA */ dns_fixedname_init(&fixed); name = dns_fixedname_name(&fixed); str = cfg_obj_asstring(cfg_tuple_get(obj, "domain")); isc_buffer_init(&b, str, strlen(str)); isc_buffer_add(&b, strlen(str)); CHECK(dns_name_fromtext(name, &b, dns_rootname, ISC_TRUE, NULL)); #endif str = cfg_obj_asstring(cfg_tuple_get(obj, "trust-anchor")); isc_buffer_init(&b, str, strlen(str)); isc_buffer_add(&b, strlen(str)); dlv = dns_fixedname_name(&view->dlv_fixed); CHECK(dns_name_fromtext(dlv, &b, dns_rootname, ISC_TRUE, NULL)); view->dlv = dns_fixedname_name(&view->dlv_fixed); } } else view->dlv = NULL; /* * For now, there is only one kind of trusted keys, the * "security roots". */ if (view->enablednssec) { CHECK(configure_view_dnsseckeys(vconfig, config, mctx, &view->secroots)); dns_resolver_resetmustbesecure(view->resolver); obj = NULL; result = ns_config_get(maps, "dnssec-must-be-secure", &obj); if (result == ISC_R_SUCCESS) CHECK(mustbesecure(obj, view->resolver)); } obj = NULL; result = ns_config_get(maps, "max-cache-ttl", &obj); INSIST(result == ISC_R_SUCCESS); view->maxcachettl = cfg_obj_asuint32(obj); obj = NULL; result = ns_config_get(maps, "max-ncache-ttl", &obj); INSIST(result == ISC_R_SUCCESS); view->maxncachettl = cfg_obj_asuint32(obj); if (view->maxncachettl > 7 * 24 * 3600) view->maxncachettl = 7 * 24 * 3600; obj = NULL; result = ns_config_get(maps, "preferred-glue", &obj); if (result == ISC_R_SUCCESS) { str = cfg_obj_asstring(obj); if (strcasecmp(str, "a") == 0) view->preferred_glue = dns_rdatatype_a; else if (strcasecmp(str, "aaaa") == 0) view->preferred_glue = dns_rdatatype_aaaa; else view->preferred_glue = 0; } else view->preferred_glue = 0; obj = NULL; result = ns_config_get(maps, "root-delegation-only", &obj); if (result == ISC_R_SUCCESS) { dns_view_setrootdelonly(view, ISC_TRUE); if (!cfg_obj_isvoid(obj)) { dns_fixedname_t fixed; dns_name_t *name; isc_buffer_t b; char *str; cfg_obj_t *exclude; dns_fixedname_init(&fixed); name = dns_fixedname_name(&fixed); for (element = cfg_list_first(obj); element != NULL; element = cfg_list_next(element)) { exclude = cfg_listelt_value(element); str = cfg_obj_asstring(exclude); isc_buffer_init(&b, str, strlen(str)); isc_buffer_add(&b, strlen(str)); CHECK(dns_name_fromtext(name, &b, dns_rootname, ISC_FALSE, NULL)); CHECK(dns_view_excludedelegationonly(view, name)); } } } else dns_view_setrootdelonly(view, ISC_FALSE); result = ISC_R_SUCCESS; cleanup: if (dispatch4 != NULL) dns_dispatch_detach(&dispatch4); if (dispatch6 != NULL) dns_dispatch_detach(&dispatch6); if (order != NULL) dns_order_detach(&order); if (cmctx != NULL) isc_mem_detach(&cmctx); if (cache != NULL) dns_cache_detach(&cache); return (result); } static isc_result_t configure_hints(dns_view_t *view, const char *filename) { isc_result_t result; dns_db_t *db; db = NULL; result = dns_rootns_create(view->mctx, view->rdclass, filename, &db); if (result == ISC_R_SUCCESS) { dns_view_sethints(view, db); dns_db_detach(&db); } return (result); } static isc_result_t configure_alternates(cfg_obj_t *config, dns_view_t *view, cfg_obj_t *alternates) { cfg_obj_t *portobj; cfg_obj_t *addresses; cfg_listelt_t *element; isc_result_t result = ISC_R_SUCCESS; in_port_t port; /* * Determine which port to send requests to. */ if (ns_g_lwresdonly && ns_g_port != 0) port = ns_g_port; else CHECKM(ns_config_getport(config, &port), "port"); if (alternates != NULL) { portobj = cfg_tuple_get(alternates, "port"); if (cfg_obj_isuint32(portobj)) { isc_uint32_t val = cfg_obj_asuint32(portobj); if (val > ISC_UINT16_MAX) { cfg_obj_log(portobj, ns_g_lctx, ISC_LOG_ERROR, "port '%u' out of range", val); return (ISC_R_RANGE); } port = (in_port_t) val; } } addresses = NULL; if (alternates != NULL) addresses = cfg_tuple_get(alternates, "addresses"); for (element = cfg_list_first(addresses); element != NULL; element = cfg_list_next(element)) { cfg_obj_t *alternate = cfg_listelt_value(element); isc_sockaddr_t sa; if (!cfg_obj_issockaddr(alternate)) { dns_fixedname_t fixed; dns_name_t *name; char *str = cfg_obj_asstring(cfg_tuple_get(alternate, "name")); isc_buffer_t buffer; in_port_t myport = port; isc_buffer_init(&buffer, str, strlen(str)); isc_buffer_add(&buffer, strlen(str)); dns_fixedname_init(&fixed); name = dns_fixedname_name(&fixed); CHECK(dns_name_fromtext(name, &buffer, dns_rootname, ISC_FALSE, NULL)); portobj = cfg_tuple_get(alternate, "port"); if (cfg_obj_isuint32(portobj)) { isc_uint32_t val = cfg_obj_asuint32(portobj); if (val > ISC_UINT16_MAX) { cfg_obj_log(portobj, ns_g_lctx, ISC_LOG_ERROR, "port '%u' out of range", val); return (ISC_R_RANGE); } myport = (in_port_t) val; } CHECK(dns_resolver_addalternate(view->resolver, NULL, name, myport)); continue; } sa = *cfg_obj_assockaddr(alternate); if (isc_sockaddr_getport(&sa) == 0) isc_sockaddr_setport(&sa, port); CHECK(dns_resolver_addalternate(view->resolver, &sa, NULL, 0)); } cleanup: return (result); } static isc_result_t configure_forward(cfg_obj_t *config, dns_view_t *view, dns_name_t *origin, cfg_obj_t *forwarders, cfg_obj_t *forwardtype) { cfg_obj_t *portobj; cfg_obj_t *faddresses; cfg_listelt_t *element; dns_fwdpolicy_t fwdpolicy = dns_fwdpolicy_none; isc_sockaddrlist_t addresses; isc_sockaddr_t *sa; isc_result_t result; in_port_t port; /* * Determine which port to send forwarded requests to. */ if (ns_g_lwresdonly && ns_g_port != 0) port = ns_g_port; else CHECKM(ns_config_getport(config, &port), "port"); if (forwarders != NULL) { portobj = cfg_tuple_get(forwarders, "port"); if (cfg_obj_isuint32(portobj)) { isc_uint32_t val = cfg_obj_asuint32(portobj); if (val > ISC_UINT16_MAX) { cfg_obj_log(portobj, ns_g_lctx, ISC_LOG_ERROR, "port '%u' out of range", val); return (ISC_R_RANGE); } port = (in_port_t) val; } } faddresses = NULL; if (forwarders != NULL) faddresses = cfg_tuple_get(forwarders, "addresses"); ISC_LIST_INIT(addresses); for (element = cfg_list_first(faddresses); element != NULL; element = cfg_list_next(element)) { cfg_obj_t *forwarder = cfg_listelt_value(element); sa = isc_mem_get(view->mctx, sizeof(isc_sockaddr_t)); if (sa == NULL) { result = ISC_R_NOMEMORY; goto cleanup; } *sa = *cfg_obj_assockaddr(forwarder); if (isc_sockaddr_getport(sa) == 0) isc_sockaddr_setport(sa, port); ISC_LINK_INIT(sa, link); ISC_LIST_APPEND(addresses, sa, link); } if (ISC_LIST_EMPTY(addresses)) { if (forwardtype != NULL) cfg_obj_log(forwarders, ns_g_lctx, ISC_LOG_WARNING, "no forwarders seen; disabling " "forwarding"); fwdpolicy = dns_fwdpolicy_none; } else { if (forwardtype == NULL) fwdpolicy = dns_fwdpolicy_first; else { char *forwardstr = cfg_obj_asstring(forwardtype); if (strcasecmp(forwardstr, "first") == 0) fwdpolicy = dns_fwdpolicy_first; else if (strcasecmp(forwardstr, "only") == 0) fwdpolicy = dns_fwdpolicy_only; else INSIST(0); } } result = dns_fwdtable_add(view->fwdtable, origin, &addresses, fwdpolicy); if (result != ISC_R_SUCCESS) { char namebuf[DNS_NAME_FORMATSIZE]; dns_name_format(origin, namebuf, sizeof(namebuf)); cfg_obj_log(forwarders, ns_g_lctx, ISC_LOG_WARNING, "could not set up forwarding for domain '%s': %s", namebuf, isc_result_totext(result)); goto cleanup; } result = ISC_R_SUCCESS; cleanup: while (!ISC_LIST_EMPTY(addresses)) { sa = ISC_LIST_HEAD(addresses); ISC_LIST_UNLINK(addresses, sa, link); isc_mem_put(view->mctx, sa, sizeof(isc_sockaddr_t)); } return (result); } /* * Create a new view and add it to the list. * * If 'vconfig' is NULL, create the default view. * * The view created is attached to '*viewp'. */ static isc_result_t create_view(cfg_obj_t *vconfig, dns_viewlist_t *viewlist, dns_view_t **viewp) { isc_result_t result; const char *viewname; dns_rdataclass_t viewclass; dns_view_t *view = NULL; if (vconfig != NULL) { cfg_obj_t *classobj = NULL; viewname = cfg_obj_asstring(cfg_tuple_get(vconfig, "name")); classobj = cfg_tuple_get(vconfig, "class"); result = ns_config_getclass(classobj, dns_rdataclass_in, &viewclass); } else { viewname = "_default"; viewclass = dns_rdataclass_in; } result = dns_viewlist_find(viewlist, viewname, viewclass, &view); if (result == ISC_R_SUCCESS) return (ISC_R_EXISTS); if (result != ISC_R_NOTFOUND) return (result); INSIST(view == NULL); result = dns_view_create(ns_g_mctx, viewclass, viewname, &view); if (result != ISC_R_SUCCESS) return (result); ISC_LIST_APPEND(*viewlist, view, link); dns_view_attach(view, viewp); return (ISC_R_SUCCESS); } /* * Configure or reconfigure a zone. */ static isc_result_t configure_zone(cfg_obj_t *config, cfg_obj_t *zconfig, cfg_obj_t *vconfig, isc_mem_t *mctx, dns_view_t *view, ns_aclconfctx_t *aclconf) { dns_view_t *pview = NULL; /* Production view */ dns_zone_t *zone = NULL; /* New or reused zone */ dns_zone_t *dupzone = NULL; cfg_obj_t *options = NULL; cfg_obj_t *zoptions = NULL; cfg_obj_t *typeobj = NULL; cfg_obj_t *forwarders = NULL; cfg_obj_t *forwardtype = NULL; cfg_obj_t *only = NULL; isc_result_t result; isc_result_t tresult; isc_buffer_t buffer; dns_fixedname_t fixorigin; dns_name_t *origin; const char *zname; dns_rdataclass_t zclass; const char *ztypestr; options = NULL; (void)cfg_map_get(config, "options", &options); zoptions = cfg_tuple_get(zconfig, "options"); /* * Get the zone origin as a dns_name_t. */ zname = cfg_obj_asstring(cfg_tuple_get(zconfig, "name")); isc_buffer_init(&buffer, zname, strlen(zname)); isc_buffer_add(&buffer, strlen(zname)); dns_fixedname_init(&fixorigin); CHECK(dns_name_fromtext(dns_fixedname_name(&fixorigin), &buffer, dns_rootname, ISC_FALSE, NULL)); origin = dns_fixedname_name(&fixorigin); CHECK(ns_config_getclass(cfg_tuple_get(zconfig, "class"), view->rdclass, &zclass)); if (zclass != view->rdclass) { const char *vname = NULL; if (vconfig != NULL) vname = cfg_obj_asstring(cfg_tuple_get(vconfig, "name")); else vname = ""; isc_log_write(ns_g_lctx, NS_LOGCATEGORY_GENERAL, NS_LOGMODULE_SERVER, ISC_LOG_ERROR, "zone '%s': wrong class for view '%s'", zname, vname); result = ISC_R_FAILURE; goto cleanup; } (void)cfg_map_get(zoptions, "type", &typeobj); if (typeobj == NULL) { cfg_obj_log(zconfig, ns_g_lctx, ISC_LOG_ERROR, "zone '%s' 'type' not specified", zname); return (ISC_R_FAILURE); } ztypestr = cfg_obj_asstring(typeobj); /* * "hints zones" aren't zones. If we've got one, * configure it and return. */ if (strcasecmp(ztypestr, "hint") == 0) { cfg_obj_t *fileobj = NULL; if (cfg_map_get(zoptions, "file", &fileobj) != ISC_R_SUCCESS) { isc_log_write(ns_g_lctx, NS_LOGCATEGORY_GENERAL, NS_LOGMODULE_SERVER, ISC_LOG_ERROR, "zone '%s': 'file' not specified", zname); result = ISC_R_FAILURE; goto cleanup; } if (dns_name_equal(origin, dns_rootname)) { char *hintsfile = cfg_obj_asstring(fileobj); result = configure_hints(view, hintsfile); if (result != ISC_R_SUCCESS) { isc_log_write(ns_g_lctx, NS_LOGCATEGORY_GENERAL, NS_LOGMODULE_SERVER, ISC_LOG_ERROR, "could not configure root hints " "from '%s': %s", hintsfile, isc_result_totext(result)); goto cleanup; } /* * Hint zones may also refer to delegation only points. */ only = NULL; tresult = cfg_map_get(zoptions, "delegation-only", &only); if (tresult == ISC_R_SUCCESS && cfg_obj_asboolean(only)) CHECK(dns_view_adddelegationonly(view, origin)); } else { isc_log_write(ns_g_lctx, NS_LOGCATEGORY_GENERAL, NS_LOGMODULE_SERVER, ISC_LOG_WARNING, "ignoring non-root hint zone '%s'", zname); result = ISC_R_SUCCESS; } /* Skip ordinary zone processing. */ goto cleanup; } /* * "forward zones" aren't zones either. Translate this syntax into * the appropriate selective forwarding configuration and return. */ if (strcasecmp(ztypestr, "forward") == 0) { forwardtype = NULL; forwarders = NULL; (void)cfg_map_get(zoptions, "forward", &forwardtype); (void)cfg_map_get(zoptions, "forwarders", &forwarders); result = configure_forward(config, view, origin, forwarders, forwardtype); goto cleanup; } /* * "delegation-only zones" aren't zones either. */ if (strcasecmp(ztypestr, "delegation-only") == 0) { result = dns_view_adddelegationonly(view, origin); goto cleanup; } /* * Check for duplicates in the new zone table. */ result = dns_view_findzone(view, origin, &dupzone); if (result == ISC_R_SUCCESS) { /* * We already have this zone! */ cfg_obj_log(zconfig, ns_g_lctx, ISC_LOG_ERROR, "zone '%s' already exists", zname); dns_zone_detach(&dupzone); result = ISC_R_EXISTS; goto cleanup; } INSIST(dupzone == NULL); /* * See if we can reuse an existing zone. This is * only possible if all of these are true: * - The zone's view exists * - A zone with the right name exists in the view * - The zone is compatible with the config * options (e.g., an existing master zone cannot * be reused if the options specify a slave zone) */ result = dns_viewlist_find(&ns_g_server->viewlist, view->name, view->rdclass, &pview); if (result != ISC_R_NOTFOUND && result != ISC_R_SUCCESS) goto cleanup; if (pview != NULL) result = dns_view_findzone(pview, origin, &zone); if (result != ISC_R_NOTFOUND && result != ISC_R_SUCCESS) goto cleanup; if (zone != NULL) { if (! ns_zone_reusable(zone, zconfig)) dns_zone_detach(&zone); } if (zone != NULL) { /* * We found a reusable zone. Make it use the * new view. */ dns_zone_setview(zone, view); } else { /* * We cannot reuse an existing zone, we have * to create a new one. */ CHECK(dns_zone_create(&zone, mctx)); CHECK(dns_zone_setorigin(zone, origin)); dns_zone_setview(zone, view); CHECK(dns_zonemgr_managezone(ns_g_server->zonemgr, zone)); } /* * If the zone contains a 'forwarders' statement, configure * selective forwarding. */ forwarders = NULL; if (cfg_map_get(zoptions, "forwarders", &forwarders) == ISC_R_SUCCESS) { forwardtype = NULL; (void)cfg_map_get(zoptions, "forward", &forwardtype); CHECK(configure_forward(config, view, origin, forwarders, forwardtype)); } /* * Stub and forward zones may also refer to delegation only points. */ only = NULL; if (cfg_map_get(zoptions, "delegation-only", &only) == ISC_R_SUCCESS) { if (cfg_obj_asboolean(only)) CHECK(dns_view_adddelegationonly(view, origin)); } /* * Configure the zone. */ CHECK(ns_zone_configure(config, vconfig, zconfig, aclconf, zone)); /* * Add the zone to its view in the new view list. */ CHECK(dns_view_addzone(view, zone)); cleanup: if (zone != NULL) dns_zone_detach(&zone); if (pview != NULL) dns_view_detach(&pview); return (result); } /* * Configure a single server quota. */ static void configure_server_quota(cfg_obj_t **maps, const char *name, isc_quota_t *quota) { cfg_obj_t *obj = NULL; isc_result_t result; result = ns_config_get(maps, name, &obj); INSIST(result == ISC_R_SUCCESS); quota->max = cfg_obj_asuint32(obj); } /* * This function is called as soon as the 'directory' statement has been * parsed. This can be extended to support other options if necessary. */ static isc_result_t directory_callback(const char *clausename, cfg_obj_t *obj, void *arg) { isc_result_t result; char *directory; REQUIRE(strcasecmp("directory", clausename) == 0); UNUSED(arg); UNUSED(clausename); /* * Change directory. */ directory = cfg_obj_asstring(obj); if (! isc_file_ischdiridempotent(directory)) cfg_obj_log(obj, ns_g_lctx, ISC_LOG_WARNING, "option 'directory' contains relative path '%s'", directory); result = isc_dir_chdir(directory); if (result != ISC_R_SUCCESS) { cfg_obj_log(obj, ns_g_lctx, ISC_LOG_ERROR, "change directory to '%s' failed: %s", directory, isc_result_totext(result)); return (result); } return (ISC_R_SUCCESS); } static void scan_interfaces(ns_server_t *server, isc_boolean_t verbose) { isc_boolean_t match_mapped = server->aclenv.match_mapped; ns_interfacemgr_scan(server->interfacemgr, verbose); /* * Update the "localhost" and "localnets" ACLs to match the * current set of network interfaces. */ dns_aclenv_copy(&server->aclenv, ns_interfacemgr_getaclenv(server->interfacemgr)); server->aclenv.match_mapped = match_mapped; } static isc_result_t add_listenelt(isc_mem_t *mctx, ns_listenlist_t *list, isc_sockaddr_t *addr) { ns_listenelt_t *lelt = NULL; dns_acl_t *src_acl = NULL; dns_aclelement_t aelt; isc_result_t result; isc_sockaddr_t any_sa6; REQUIRE(isc_sockaddr_pf(addr) == AF_INET6); isc_sockaddr_any6(&any_sa6); if (!isc_sockaddr_equal(&any_sa6, addr)) { aelt.type = dns_aclelementtype_ipprefix; aelt.negative = ISC_FALSE; aelt.u.ip_prefix.prefixlen = 128; isc_netaddr_fromin6(&aelt.u.ip_prefix.address, &addr->type.sin6.sin6_addr); result = dns_acl_create(mctx, 1, &src_acl); if (result != ISC_R_SUCCESS) return (result); result = dns_acl_appendelement(src_acl, &aelt); if (result != ISC_R_SUCCESS) goto clean; result = ns_listenelt_create(mctx, isc_sockaddr_getport(addr), src_acl, &lelt); if (result != ISC_R_SUCCESS) goto clean; ISC_LIST_APPEND(list->elts, lelt, link); } return (ISC_R_SUCCESS); clean: INSIST(lelt == NULL); if (src_acl != NULL) dns_acl_detach(&src_acl); return (result); } /* * Make a list of xxx-source addresses and call ns_interfacemgr_adjust() * to update the listening interfaces accordingly. * We currently only consider IPv6, because this only affects IPv6 wildcard * sockets. */ static void adjust_interfaces(ns_server_t *server, isc_mem_t *mctx) { isc_result_t result; ns_listenlist_t *list = NULL; dns_view_t *view; dns_zone_t *zone, *next; isc_sockaddr_t addr, *addrp; result = ns_listenlist_create(mctx, &list); if (result != ISC_R_SUCCESS) return; for (view = ISC_LIST_HEAD(server->viewlist); view != NULL; view = ISC_LIST_NEXT(view, link)) { dns_dispatch_t *dispatch6; dispatch6 = dns_resolver_dispatchv6(view->resolver); if (dispatch6 == NULL) continue; result = dns_dispatch_getlocaladdress(dispatch6, &addr); if (result != ISC_R_SUCCESS) goto fail; result = add_listenelt(mctx, list, &addr); if (result != ISC_R_SUCCESS) goto fail; } zone = NULL; for (result = dns_zone_first(server->zonemgr, &zone); result == ISC_R_SUCCESS; next = NULL, result = dns_zone_next(zone, &next), zone = next) { dns_view_t *zoneview; /* * At this point the zone list may contain a stale zone * just removed from the configuration. To see the validity, * check if the corresponding view is in our current view list. */ zoneview = dns_zone_getview(zone); INSIST(zoneview != NULL); for (view = ISC_LIST_HEAD(server->viewlist); view != NULL && view != zoneview; view = ISC_LIST_NEXT(view, link)) ; if (view == NULL) continue; addrp = dns_zone_getnotifysrc6(zone); result = add_listenelt(mctx, list, addrp); if (result != ISC_R_SUCCESS) goto fail; addrp = dns_zone_getxfrsource6(zone); result = add_listenelt(mctx, list, addrp); if (result != ISC_R_SUCCESS) goto fail; } ns_interfacemgr_adjust(server->interfacemgr, list, ISC_TRUE); clean: ns_listenlist_detach(&list); return; fail: /* * Even when we failed the procedure, most of other interfaces * should work correctly. We therefore just warn it. */ isc_log_write(ns_g_lctx, NS_LOGCATEGORY_GENERAL, NS_LOGMODULE_SERVER, ISC_LOG_WARNING, "could not adjust the listen-on list; " "some interfaces may not work"); goto clean; } /* * This event callback is invoked to do periodic network * interface scanning. */ static void interface_timer_tick(isc_task_t *task, isc_event_t *event) { isc_result_t result; ns_server_t *server = (ns_server_t *) event->ev_arg; INSIST(task == server->task); UNUSED(task); isc_event_free(&event); /* * XXX should scan interfaces unlocked and get exclusive access * only to replace ACLs. */ result = isc_task_beginexclusive(server->task); RUNTIME_CHECK(result == ISC_R_SUCCESS); scan_interfaces(server, ISC_FALSE); isc_task_endexclusive(server->task); } static void heartbeat_timer_tick(isc_task_t *task, isc_event_t *event) { ns_server_t *server = (ns_server_t *) event->ev_arg; dns_view_t *view; UNUSED(task); isc_event_free(&event); view = ISC_LIST_HEAD(server->viewlist); while (view != NULL) { dns_view_dialup(view); view = ISC_LIST_NEXT(view, link); } } /* * Replace the current value of '*field', a dynamically allocated * string or NULL, with a dynamically allocated copy of the * null-terminated string pointed to by 'value', or NULL. */ static isc_result_t setstring(ns_server_t *server, char **field, const char *value) { char *copy; if (value != NULL) { copy = isc_mem_strdup(server->mctx, value); if (copy == NULL) return (ISC_R_NOMEMORY); } else { copy = NULL; } if (*field != NULL) isc_mem_free(server->mctx, *field); *field = copy; return (ISC_R_SUCCESS); } /* * Replace the current value of '*field', a dynamically allocated * string or NULL, with another dynamically allocated string * or NULL if whether 'obj' is a string or void value, respectively. */ static isc_result_t setoptstring(ns_server_t *server, char **field, cfg_obj_t *obj) { if (cfg_obj_isvoid(obj)) return (setstring(server, field, NULL)); else return (setstring(server, field, cfg_obj_asstring(obj))); } static void set_limit(cfg_obj_t **maps, const char *configname, const char *description, isc_resource_t resourceid, isc_resourcevalue_t defaultvalue) { cfg_obj_t *obj = NULL; char *resource; isc_resourcevalue_t value; isc_result_t result; if (ns_config_get(maps, configname, &obj) != ISC_R_SUCCESS) return; if (cfg_obj_isstring(obj)) { resource = cfg_obj_asstring(obj); if (strcasecmp(resource, "unlimited") == 0) value = ISC_RESOURCE_UNLIMITED; else { INSIST(strcasecmp(resource, "default") == 0); value = defaultvalue; } } else value = cfg_obj_asuint64(obj); result = isc_resource_setlimit(resourceid, value); isc_log_write(ns_g_lctx, NS_LOGCATEGORY_GENERAL, NS_LOGMODULE_SERVER, result == ISC_R_SUCCESS ? ISC_LOG_DEBUG(3) : ISC_LOG_WARNING, "set maximum %s to %" ISC_PRINT_QUADFORMAT "d: %s", description, value, isc_result_totext(result)); } #define SETLIMIT(cfgvar, resource, description) \ set_limit(maps, cfgvar, description, isc_resource_ ## resource, \ ns_g_init ## resource) static void set_limits(cfg_obj_t **maps) { SETLIMIT("stacksize", stacksize, "stack size"); SETLIMIT("datasize", datasize, "data size"); SETLIMIT("coresize", coresize, "core size"); SETLIMIT("files", openfiles, "open files"); } static isc_result_t portlist_fromconf(dns_portlist_t *portlist, unsigned int family, cfg_obj_t *ports) { cfg_listelt_t *element; isc_result_t result = ISC_R_SUCCESS; for (element = cfg_list_first(ports); element != NULL; element = cfg_list_next(element)) { cfg_obj_t *obj = cfg_listelt_value(element); in_port_t port = (in_port_t)cfg_obj_asuint32(obj); result = dns_portlist_add(portlist, family, port); if (result != ISC_R_SUCCESS) break; } return (result); } static isc_result_t load_configuration(const char *filename, ns_server_t *server, isc_boolean_t first_time) { isc_result_t result; cfg_parser_t *parser = NULL; cfg_obj_t *config; cfg_obj_t *options; cfg_obj_t *views; cfg_obj_t *obj; cfg_obj_t *v4ports, *v6ports; cfg_obj_t *maps[3]; cfg_obj_t *builtin_views; cfg_listelt_t *element; dns_view_t *view = NULL; dns_view_t *view_next; dns_viewlist_t viewlist; dns_viewlist_t tmpviewlist; ns_aclconfctx_t aclconfctx; isc_uint32_t interface_interval; isc_uint32_t heartbeat_interval; isc_uint32_t udpsize; in_port_t listen_port; int i; ns_aclconfctx_init(&aclconfctx); ISC_LIST_INIT(viewlist); /* Ensure exclusive access to configuration data. */ result = isc_task_beginexclusive(server->task); RUNTIME_CHECK(result == ISC_R_SUCCESS); /* * Parse the global default pseudo-config file. */ if (first_time) { CHECK(ns_config_parsedefaults(ns_g_parser, &ns_g_config)); RUNTIME_CHECK(cfg_map_get(ns_g_config, "options", &ns_g_defaults) == ISC_R_SUCCESS); } /* * Parse the configuration file using the new config code. */ result = ISC_R_FAILURE; config = NULL; /* * Unless this is lwresd with the -C option, parse the config file. */ if (!(ns_g_lwresdonly && lwresd_g_useresolvconf)) { isc_log_write(ns_g_lctx, NS_LOGCATEGORY_GENERAL, NS_LOGMODULE_SERVER, ISC_LOG_INFO, "loading configuration from '%s'", filename); CHECK(cfg_parser_create(ns_g_mctx, ns_g_lctx, &parser)); cfg_parser_setcallback(parser, directory_callback, NULL); result = cfg_parse_file(parser, filename, &cfg_type_namedconf, &config); } /* * If this is lwresd with the -C option, or lwresd with no -C or -c * option where the above parsing failed, parse resolv.conf. */ if (ns_g_lwresdonly && (lwresd_g_useresolvconf || (!ns_g_conffileset && result == ISC_R_FILENOTFOUND))) { isc_log_write(ns_g_lctx, NS_LOGCATEGORY_GENERAL, NS_LOGMODULE_SERVER, ISC_LOG_INFO, "loading configuration from '%s'", lwresd_g_resolvconffile); if (parser != NULL) cfg_parser_destroy(&parser); CHECK(cfg_parser_create(ns_g_mctx, ns_g_lctx, &parser)); result = ns_lwresd_parseeresolvconf(ns_g_mctx, parser, &config); } CHECK(result); /* * Check the validity of the configuration. */ CHECK(bind9_check_namedconf(config, ns_g_lctx, ns_g_mctx)); /* * Fill in the maps array, used for resolving defaults. */ i = 0; options = NULL; result = cfg_map_get(config, "options", &options); if (result == ISC_R_SUCCESS) maps[i++] = options; maps[i++] = ns_g_defaults; maps[i++] = NULL; /* * Set process limits, which (usually) needs to be done as root. */ set_limits(maps); /* * Configure various server options. */ configure_server_quota(maps, "transfers-out", &server->xfroutquota); configure_server_quota(maps, "tcp-clients", &server->tcpquota); configure_server_quota(maps, "recursive-clients", &server->recursionquota); CHECK(configure_view_acl(NULL, config, "blackhole", &aclconfctx, ns_g_mctx, &server->blackholeacl)); if (server->blackholeacl != NULL) dns_dispatchmgr_setblackhole(ns_g_dispatchmgr, server->blackholeacl); obj = NULL; result = ns_config_get(maps, "match-mapped-addresses", &obj); INSIST(result == ISC_R_SUCCESS); server->aclenv.match_mapped = cfg_obj_asboolean(obj); v4ports = NULL; v6ports = NULL; (void)ns_config_get(maps, "avoid-v4-udp-ports", &v4ports); (void)ns_config_get(maps, "avoid-v6-udp-ports", &v6ports); if (v4ports != NULL || v6ports != NULL) { dns_portlist_t *portlist = NULL; result = dns_portlist_create(ns_g_mctx, &portlist); if (result == ISC_R_SUCCESS && v4ports != NULL) result = portlist_fromconf(portlist, AF_INET, v4ports); if (result == ISC_R_SUCCESS && v6ports != NULL) portlist_fromconf(portlist, AF_INET6, v6ports); if (result == ISC_R_SUCCESS) dns_dispatchmgr_setblackportlist(ns_g_dispatchmgr, portlist); if (portlist != NULL) dns_portlist_detach(&portlist); CHECK(result); } else dns_dispatchmgr_setblackportlist(ns_g_dispatchmgr, NULL); /* * Set the EDNS UDP size when we don't match a view. */ obj = NULL; result = ns_config_get(maps, "edns-udp-size", &obj); INSIST(result == ISC_R_SUCCESS); udpsize = cfg_obj_asuint32(obj); if (udpsize < 512) udpsize = 512; if (udpsize > 4096) udpsize = 4096; ns_g_udpsize = (isc_uint16_t)udpsize; /* * Configure the zone manager. */ obj = NULL; result = ns_config_get(maps, "transfers-in", &obj); INSIST(result == ISC_R_SUCCESS); dns_zonemgr_settransfersin(server->zonemgr, cfg_obj_asuint32(obj)); obj = NULL; result = ns_config_get(maps, "transfers-per-ns", &obj); INSIST(result == ISC_R_SUCCESS); dns_zonemgr_settransfersperns(server->zonemgr, cfg_obj_asuint32(obj)); obj = NULL; result = ns_config_get(maps, "serial-query-rate", &obj); INSIST(result == ISC_R_SUCCESS); dns_zonemgr_setserialqueryrate(server->zonemgr, cfg_obj_asuint32(obj)); /* * Determine which port to use for listening for incoming connections. */ if (ns_g_port != 0) listen_port = ns_g_port; else CHECKM(ns_config_getport(config, &listen_port), "port"); /* * Find the listen queue depth. */ obj = NULL; result = ns_config_get(maps, "tcp-listen-queue", &obj); INSIST(result == ISC_R_SUCCESS); ns_g_listen = cfg_obj_asuint32(obj); if (ns_g_listen < 3) ns_g_listen = 3; /* * Configure the interface manager according to the "listen-on" * statement. */ { cfg_obj_t *clistenon = NULL; ns_listenlist_t *listenon = NULL; clistenon = NULL; /* * Even though listen-on is present in the default * configuration, we can't use it here, since it isn't * used if we're in lwresd mode. This way is easier. */ if (options != NULL) (void)cfg_map_get(options, "listen-on", &clistenon); if (clistenon != NULL) { result = ns_listenlist_fromconfig(clistenon, config, &aclconfctx, ns_g_mctx, &listenon); } else if (!ns_g_lwresdonly) { /* * Not specified, use default. */ CHECK(ns_listenlist_default(ns_g_mctx, listen_port, ISC_TRUE, &listenon)); } if (listenon != NULL) { ns_interfacemgr_setlistenon4(server->interfacemgr, listenon); ns_listenlist_detach(&listenon); } } /* * Ditto for IPv6. */ { cfg_obj_t *clistenon = NULL; ns_listenlist_t *listenon = NULL; if (options != NULL) (void)cfg_map_get(options, "listen-on-v6", &clistenon); if (clistenon != NULL) { result = ns_listenlist_fromconfig(clistenon, config, &aclconfctx, ns_g_mctx, &listenon); } else if (!ns_g_lwresdonly) { /* * Not specified, use default. */ CHECK(ns_listenlist_default(ns_g_mctx, listen_port, ISC_TRUE, &listenon)); } if (listenon != NULL) { ns_interfacemgr_setlistenon6(server->interfacemgr, listenon); ns_listenlist_detach(&listenon); } } /* * Rescan the interface list to pick up changes in the * listen-on option. It's important that we do this before we try * to configure the query source, since the dispatcher we use might * be shared with an interface. */ scan_interfaces(server, ISC_TRUE); /* * Arrange for further interface scanning to occur periodically * as specified by the "interface-interval" option. */ obj = NULL; result = ns_config_get(maps, "interface-interval", &obj); INSIST(result == ISC_R_SUCCESS); interface_interval = cfg_obj_asuint32(obj) * 60; if (interface_interval == 0) { CHECK(isc_timer_reset(server->interface_timer, isc_timertype_inactive, NULL, NULL, ISC_TRUE)); } else if (server->interface_interval != interface_interval) { isc_interval_t interval; isc_interval_set(&interval, interface_interval, 0); CHECK(isc_timer_reset(server->interface_timer, isc_timertype_ticker, NULL, &interval, ISC_FALSE)); } server->interface_interval = interface_interval; /* * Configure the dialup heartbeat timer. */ obj = NULL; result = ns_config_get(maps, "heartbeat-interval", &obj); INSIST(result == ISC_R_SUCCESS); heartbeat_interval = cfg_obj_asuint32(obj) * 60; if (heartbeat_interval == 0) { CHECK(isc_timer_reset(server->heartbeat_timer, isc_timertype_inactive, NULL, NULL, ISC_TRUE)); } else if (server->heartbeat_interval != heartbeat_interval) { isc_interval_t interval; isc_interval_set(&interval, heartbeat_interval, 0); CHECK(isc_timer_reset(server->heartbeat_timer, isc_timertype_ticker, NULL, &interval, ISC_FALSE)); } server->heartbeat_interval = heartbeat_interval; /* * Configure and freeze all explicit views. Explicit * views that have zones were already created at parsing * time, but views with no zones must be created here. */ views = NULL; (void)cfg_map_get(config, "view", &views); for (element = cfg_list_first(views); element != NULL; element = cfg_list_next(element)) { cfg_obj_t *vconfig = cfg_listelt_value(element); view = NULL; CHECK(create_view(vconfig, &viewlist, &view)); INSIST(view != NULL); CHECK(configure_view(view, config, vconfig, ns_g_mctx, &aclconfctx, ISC_TRUE)); dns_view_freeze(view); dns_view_detach(&view); } /* * Make sure we have a default view if and only if there * were no explicit views. */ if (views == NULL) { /* * No explicit views; there ought to be a default view. * There may already be one created as a side effect * of zone statements, or we may have to create one. * In either case, we need to configure and freeze it. */ CHECK(create_view(NULL, &viewlist, &view)); CHECK(configure_view(view, config, NULL, ns_g_mctx, &aclconfctx, ISC_TRUE)); dns_view_freeze(view); dns_view_detach(&view); } /* * Create (or recreate) the built-in views. Currently * there is only one, the _bind view. */ builtin_views = NULL; RUNTIME_CHECK(cfg_map_get(ns_g_config, "view", &builtin_views) == ISC_R_SUCCESS); for (element = cfg_list_first(builtin_views); element != NULL; element = cfg_list_next(element)) { cfg_obj_t *vconfig = cfg_listelt_value(element); CHECK(create_view(vconfig, &viewlist, &view)); CHECK(configure_view(view, config, vconfig, ns_g_mctx, &aclconfctx, ISC_FALSE)); dns_view_freeze(view); dns_view_detach(&view); view = NULL; } /* * Swap our new view list with the production one. */ tmpviewlist = server->viewlist; server->viewlist = viewlist; viewlist = tmpviewlist; /* * Load the TKEY information from the configuration. */ if (options != NULL) { dns_tkeyctx_t *t = NULL; CHECKM(ns_tkeyctx_fromconfig(options, ns_g_mctx, ns_g_entropy, &t), "configuring TKEY"); if (server->tkeyctx != NULL) dns_tkeyctx_destroy(&server->tkeyctx); server->tkeyctx = t; } /* * Bind the control port(s). */ CHECKM(ns_controls_configure(ns_g_server->controls, config, &aclconfctx), "binding control channel(s)"); /* * Bind the lwresd port(s). */ CHECKM(ns_lwresd_configure(ns_g_mctx, config), "binding lightweight resolver ports"); /* * Open the source of entropy. */ if (first_time) { obj = NULL; result = ns_config_get(maps, "random-device", &obj); if (result != ISC_R_SUCCESS) { isc_log_write(ns_g_lctx, NS_LOGCATEGORY_GENERAL, NS_LOGMODULE_SERVER, ISC_LOG_INFO, "no source of entropy found"); } else { const char *randomdev = cfg_obj_asstring(obj); result = isc_entropy_createfilesource(ns_g_entropy, randomdev); if (result != ISC_R_SUCCESS) isc_log_write(ns_g_lctx, NS_LOGCATEGORY_GENERAL, NS_LOGMODULE_SERVER, ISC_LOG_INFO, "could not open entropy source " "%s: %s", randomdev, isc_result_totext(result)); #ifdef PATH_RANDOMDEV if (ns_g_fallbackentropy != NULL) { if (result != ISC_R_SUCCESS) { isc_log_write(ns_g_lctx, NS_LOGCATEGORY_GENERAL, NS_LOGMODULE_SERVER, ISC_LOG_INFO, "using pre-chroot entropy source " "%s", PATH_RANDOMDEV); isc_entropy_detach(&ns_g_entropy); isc_entropy_attach(ns_g_fallbackentropy, &ns_g_entropy); } isc_entropy_detach(&ns_g_fallbackentropy); } #endif } } /* * Relinquish root privileges. Not used due to privsep */ #if 0 if (first_time) ns_os_changeuser(); #endif /* * Configure the logging system. * * Do this after changing UID to make sure that any log * files specified in named.conf get created by the * unprivileged user, not root. */ if (ns_g_logstderr) { isc_log_write(ns_g_lctx, NS_LOGCATEGORY_GENERAL, NS_LOGMODULE_SERVER, ISC_LOG_INFO, "ignoring config file logging " "statement due to -g option"); } else { cfg_obj_t *logobj = NULL; isc_logconfig_t *logc = NULL; CHECKM(isc_logconfig_create(ns_g_lctx, &logc), "creating new logging configuration"); logobj = NULL; (void)cfg_map_get(config, "logging", &logobj); if (logobj != NULL) { CHECKM(ns_log_configure(logc, logobj), "configuring logging"); } else { CHECKM(ns_log_setdefaultchannels(logc), "setting up default logging channels"); CHECKM(ns_log_setunmatchedcategory(logc), "setting up default 'category unmatched'"); CHECKM(ns_log_setdefaultcategory(logc), "setting up default 'category default'"); } result = isc_logconfig_use(ns_g_lctx, logc); if (result != ISC_R_SUCCESS) { isc_logconfig_destroy(&logc); CHECKM(result, "installing logging configuration"); } isc_log_write(ns_g_lctx, NS_LOGCATEGORY_GENERAL, NS_LOGMODULE_SERVER, ISC_LOG_DEBUG(1), "now using logging configuration from " "config file"); } /* * Set the default value of the query logging flag depending * whether a "queries" category has been defined. This is * a disgusting hack, but we need to do this for BIND 8 * compatibility. */ if (first_time) { cfg_obj_t *logobj = NULL; cfg_obj_t *categories = NULL; obj = NULL; if (ns_config_get(maps, "querylog", &obj) == ISC_R_SUCCESS) { server->log_queries = cfg_obj_asboolean(obj); } else { (void)cfg_map_get(config, "logging", &logobj); if (logobj != NULL) (void)cfg_map_get(logobj, "category", &categories); if (categories != NULL) { cfg_listelt_t *element; for (element = cfg_list_first(categories); element != NULL; element = cfg_list_next(element)) { cfg_obj_t *catobj; char *str; obj = cfg_listelt_value(element); catobj = cfg_tuple_get(obj, "name"); str = cfg_obj_asstring(catobj); if (strcasecmp(str, "queries") == 0) server->log_queries = ISC_TRUE; } } } } if (ns_g_pidfile != NULL) { ns_os_writepidfile(ns_g_pidfile, first_time); } else { obj = NULL; if (ns_config_get(maps, "pid-file", &obj) == ISC_R_SUCCESS) ns_os_writepidfile(cfg_obj_asstring(obj), first_time); else if (ns_g_lwresdonly) ns_os_writepidfile(lwresd_g_defaultpidfile, first_time); else ns_os_writepidfile(ns_g_defaultpidfile, first_time); } obj = NULL; if (options != NULL && cfg_map_get(options, "memstatistics-file", &obj) == ISC_R_SUCCESS) ns_main_setmemstats(cfg_obj_asstring(obj)); else ns_main_setmemstats(NULL); obj = NULL; result = ns_config_get(maps, "statistics-file", &obj); INSIST(result == ISC_R_SUCCESS); CHECKM(setstring(server, &server->statsfile, cfg_obj_asstring(obj)), "strdup"); obj = NULL; result = ns_config_get(maps, "dump-file", &obj); INSIST(result == ISC_R_SUCCESS); CHECKM(setstring(server, &server->dumpfile, cfg_obj_asstring(obj)), "strdup"); obj = NULL; result = ns_config_get(maps, "recursing-file", &obj); INSIST(result == ISC_R_SUCCESS); CHECKM(setstring(server, &server->recfile, cfg_obj_asstring(obj)), "strdup"); obj = NULL; result = ns_config_get(maps, "version", &obj); if (result == ISC_R_SUCCESS) { CHECKM(setoptstring(server, &server->version, obj), "strdup"); server->version_set = ISC_TRUE; } else { server->version_set = ISC_FALSE; } obj = NULL; result = ns_config_get(maps, "hostname", &obj); if (result == ISC_R_SUCCESS) { CHECKM(setoptstring(server, &server->hostname, obj), "strdup"); server->hostname_set = ISC_TRUE; } else { server->hostname_set = ISC_FALSE; } obj = NULL; result = ns_config_get(maps, "server-id", &obj); server->server_usehostname = ISC_FALSE; if (result == ISC_R_SUCCESS && cfg_obj_isboolean(obj)) { server->server_usehostname = ISC_TRUE; } else if (result == ISC_R_SUCCESS) { CHECKM(setoptstring(server, &server->server_id, obj), "strdup"); } else { result = setoptstring(server, &server->server_id, NULL); RUNTIME_CHECK(result == ISC_R_SUCCESS); } obj = NULL; result = ns_config_get(maps, "flush-zones-on-shutdown", &obj); if (result == ISC_R_SUCCESS) { server->flushonshutdown = cfg_obj_asboolean(obj); } else { server->flushonshutdown = ISC_FALSE; } result = ISC_R_SUCCESS; cleanup: ns_aclconfctx_destroy(&aclconfctx); if (parser != NULL) { if (config != NULL) cfg_obj_destroy(parser, &config); cfg_parser_destroy(&parser); } if (view != NULL) dns_view_detach(&view); /* * This cleans up either the old production view list * or our temporary list depending on whether they * were swapped above or not. */ for (view = ISC_LIST_HEAD(viewlist); view != NULL; view = view_next) { view_next = ISC_LIST_NEXT(view, link); ISC_LIST_UNLINK(viewlist, view, link); dns_view_detach(&view); } /* * Adjust the listening interfaces in accordance with the source * addresses specified in views and zones. */ if (isc_net_probeipv6() == ISC_R_SUCCESS) adjust_interfaces(server, ns_g_mctx); /* Relinquish exclusive access to configuration data. */ isc_task_endexclusive(server->task); isc_log_write(ns_g_lctx, NS_LOGCATEGORY_GENERAL, NS_LOGMODULE_SERVER, ISC_LOG_DEBUG(1), "load_configuration: %s", isc_result_totext(result)); return (result); } static isc_result_t load_zones(ns_server_t *server, isc_boolean_t stop) { isc_result_t result; dns_view_t *view; result = isc_task_beginexclusive(server->task); RUNTIME_CHECK(result == ISC_R_SUCCESS); /* * Load zone data from disk. */ for (view = ISC_LIST_HEAD(server->viewlist); view != NULL; view = ISC_LIST_NEXT(view, link)) { CHECK(dns_view_load(view, stop)); } /* * Force zone maintenance. Do this after loading * so that we know when we need to force AXFR of * slave zones whose master files are missing. */ CHECK(dns_zonemgr_forcemaint(server->zonemgr)); cleanup: isc_task_endexclusive(server->task); return (result); } static isc_result_t load_new_zones(ns_server_t *server, isc_boolean_t stop) { isc_result_t result; dns_view_t *view; result = isc_task_beginexclusive(server->task); RUNTIME_CHECK(result == ISC_R_SUCCESS); /* * Load zone data from disk. */ for (view = ISC_LIST_HEAD(server->viewlist); view != NULL; view = ISC_LIST_NEXT(view, link)) { CHECK(dns_view_loadnew(view, stop)); } /* * Force zone maintenance. Do this after loading * so that we know when we need to force AXFR of * slave zones whose master files are missing. */ dns_zonemgr_resumexfrs(server->zonemgr); cleanup: isc_task_endexclusive(server->task); return (result); } static void run_server(isc_task_t *task, isc_event_t *event) { isc_result_t result; ns_server_t *server = (ns_server_t *)event->ev_arg; INSIST(task == server->task); isc_event_free(&event); CHECKFATAL(dns_dispatchmgr_create(ns_g_mctx, ns_g_entropy, &ns_g_dispatchmgr), "creating dispatch manager"); CHECKFATAL(ns_interfacemgr_create(ns_g_mctx, ns_g_taskmgr, ns_g_socketmgr, ns_g_dispatchmgr, &server->interfacemgr), "creating interface manager"); CHECKFATAL(isc_timer_create(ns_g_timermgr, isc_timertype_inactive, NULL, NULL, server->task, interface_timer_tick, server, &server->interface_timer), "creating interface timer"); CHECKFATAL(isc_timer_create(ns_g_timermgr, isc_timertype_inactive, NULL, NULL, server->task, heartbeat_timer_tick, server, &server->heartbeat_timer), "creating heartbeat timer"); CHECKFATAL(cfg_parser_create(ns_g_mctx, NULL, &ns_g_parser), "creating default configuration parser"); if (ns_g_lwresdonly) CHECKFATAL(load_configuration(lwresd_g_conffile, server, ISC_TRUE), "loading configuration"); else CHECKFATAL(load_configuration(ns_g_conffile, server, ISC_TRUE), "loading configuration"); isc_hash_init(); CHECKFATAL(load_zones(server, ISC_FALSE), "loading zones"); ns_os_started(); isc_log_write(ns_g_lctx, NS_LOGCATEGORY_GENERAL, NS_LOGMODULE_SERVER, ISC_LOG_NOTICE, "running"); } void ns_server_flushonshutdown(ns_server_t *server, isc_boolean_t flush) { REQUIRE(NS_SERVER_VALID(server)); server->flushonshutdown = flush; } static void shutdown_server(isc_task_t *task, isc_event_t *event) { isc_result_t result; dns_view_t *view, *view_next; ns_server_t *server = (ns_server_t *)event->ev_arg; isc_boolean_t flush = server->flushonshutdown; UNUSED(task); INSIST(task == server->task); result = isc_task_beginexclusive(server->task); RUNTIME_CHECK(result == ISC_R_SUCCESS); isc_log_write(ns_g_lctx, NS_LOGCATEGORY_GENERAL, NS_LOGMODULE_SERVER, ISC_LOG_INFO, "shutting down%s", flush ? ": flushing changes" : ""); ns_controls_shutdown(server->controls); end_reserved_dispatches(server, ISC_TRUE); cfg_obj_destroy(ns_g_parser, &ns_g_config); cfg_parser_destroy(&ns_g_parser); for (view = ISC_LIST_HEAD(server->viewlist); view != NULL; view = view_next) { view_next = ISC_LIST_NEXT(view, link); ISC_LIST_UNLINK(server->viewlist, view, link); if (flush) dns_view_flushanddetach(&view); else dns_view_detach(&view); } isc_timer_detach(&server->interface_timer); isc_timer_detach(&server->heartbeat_timer); ns_interfacemgr_shutdown(server->interfacemgr); ns_interfacemgr_detach(&server->interfacemgr); dns_dispatchmgr_destroy(&ns_g_dispatchmgr); dns_zonemgr_shutdown(server->zonemgr); if (server->blackholeacl != NULL) dns_acl_detach(&server->blackholeacl); dns_db_detach(&server->in_roothints); isc_task_endexclusive(server->task); isc_task_detach(&server->task); isc_event_free(&event); } void ns_server_create(isc_mem_t *mctx, ns_server_t **serverp) { isc_result_t result; ns_server_t *server = isc_mem_get(mctx, sizeof(*server)); if (server == NULL) fatal("allocating server object", ISC_R_NOMEMORY); server->mctx = mctx; server->task = NULL; /* Initialize configuration data with default values. */ result = isc_quota_init(&server->xfroutquota, 10); RUNTIME_CHECK(result == ISC_R_SUCCESS); result = isc_quota_init(&server->tcpquota, 10); RUNTIME_CHECK(result == ISC_R_SUCCESS); result = isc_quota_init(&server->recursionquota, 100); RUNTIME_CHECK(result == ISC_R_SUCCESS); isc_quota_soft(&server->recursionquota, ISC_FALSE); result = dns_aclenv_init(mctx, &server->aclenv); RUNTIME_CHECK(result == ISC_R_SUCCESS); /* Initialize server data structures. */ server->zonemgr = NULL; server->interfacemgr = NULL; ISC_LIST_INIT(server->viewlist); server->in_roothints = NULL; server->blackholeacl = NULL; CHECKFATAL(dns_rootns_create(mctx, dns_rdataclass_in, NULL, &server->in_roothints), "setting up root hints"); CHECKFATAL(isc_mutex_init(&server->reload_event_lock), "initializing reload event lock"); server->reload_event = isc_event_allocate(ns_g_mctx, server, NS_EVENT_RELOAD, ns_server_reload, server, sizeof(isc_event_t)); CHECKFATAL(server->reload_event == NULL ? ISC_R_NOMEMORY : ISC_R_SUCCESS, "allocating reload event"); CHECKFATAL(dst_lib_init(ns_g_mctx, ns_g_entropy, ISC_ENTROPY_GOODONLY), "initializing DST"); server->tkeyctx = NULL; CHECKFATAL(dns_tkeyctx_create(ns_g_mctx, ns_g_entropy, &server->tkeyctx), "creating TKEY context"); /* * Setup the server task, which is responsible for coordinating * startup and shutdown of the server. */ CHECKFATAL(isc_task_create(ns_g_taskmgr, 0, &server->task), "creating server task"); isc_task_setname(server->task, "server", server); CHECKFATAL(isc_task_onshutdown(server->task, shutdown_server, server), "isc_task_onshutdown"); CHECKFATAL(isc_app_onrun(ns_g_mctx, server->task, run_server, server), "isc_app_onrun"); server->interface_timer = NULL; server->heartbeat_timer = NULL; server->interface_interval = 0; server->heartbeat_interval = 0; CHECKFATAL(dns_zonemgr_create(ns_g_mctx, ns_g_taskmgr, ns_g_timermgr, ns_g_socketmgr, &server->zonemgr), "dns_zonemgr_create"); server->statsfile = isc_mem_strdup(server->mctx, "named.stats"); CHECKFATAL(server->statsfile == NULL ? ISC_R_NOMEMORY : ISC_R_SUCCESS, "isc_mem_strdup"); server->querystats = NULL; server->dumpfile = isc_mem_strdup(server->mctx, "named_dump.db"); CHECKFATAL(server->dumpfile == NULL ? ISC_R_NOMEMORY : ISC_R_SUCCESS, "isc_mem_strdup"); server->recfile = isc_mem_strdup(server->mctx, "named.recursing"); CHECKFATAL(server->recfile == NULL ? ISC_R_NOMEMORY : ISC_R_SUCCESS, "isc_mem_strdup"); server->hostname_set = ISC_FALSE; server->hostname = NULL; server->version_set = ISC_FALSE; server->version = NULL; server->server_usehostname = ISC_FALSE; server->server_id = NULL; CHECKFATAL(dns_stats_alloccounters(ns_g_mctx, &server->querystats), "dns_stats_alloccounters"); server->flushonshutdown = ISC_FALSE; server->log_queries = ISC_FALSE; server->controls = NULL; CHECKFATAL(ns_controls_create(server, &server->controls), "ns_controls_create"); server->dispatchgen = 0; ISC_LIST_INIT(server->dispatches); server->magic = NS_SERVER_MAGIC; *serverp = server; } void ns_server_destroy(ns_server_t **serverp) { ns_server_t *server = *serverp; REQUIRE(NS_SERVER_VALID(server)); ns_controls_destroy(&server->controls); dns_stats_freecounters(server->mctx, &server->querystats); isc_mem_free(server->mctx, server->statsfile); isc_mem_free(server->mctx, server->dumpfile); isc_mem_free(server->mctx, server->recfile); if (server->version != NULL) isc_mem_free(server->mctx, server->version); if (server->hostname != NULL) isc_mem_free(server->mctx, server->hostname); if (server->server_id != NULL) isc_mem_free(server->mctx, server->server_id); dns_zonemgr_detach(&server->zonemgr); if (server->tkeyctx != NULL) dns_tkeyctx_destroy(&server->tkeyctx); dst_lib_destroy(); isc_event_free(&server->reload_event); INSIST(ISC_LIST_EMPTY(server->viewlist)); dns_aclenv_destroy(&server->aclenv); isc_quota_destroy(&server->recursionquota); isc_quota_destroy(&server->tcpquota); isc_quota_destroy(&server->xfroutquota); server->magic = 0; isc_mem_put(server->mctx, server, sizeof(*server)); *serverp = NULL; } static void fatal(const char *msg, isc_result_t result) { isc_log_write(ns_g_lctx, NS_LOGCATEGORY_GENERAL, NS_LOGMODULE_SERVER, ISC_LOG_CRITICAL, "%s: %s", msg, isc_result_totext(result)); isc_log_write(ns_g_lctx, NS_LOGCATEGORY_GENERAL, NS_LOGMODULE_SERVER, ISC_LOG_CRITICAL, "exiting (due to fatal error)"); exit(1); } static void start_reserved_dispatches(ns_server_t *server) { REQUIRE(NS_SERVER_VALID(server)); server->dispatchgen++; } static void end_reserved_dispatches(ns_server_t *server, isc_boolean_t all) { ns_dispatch_t *dispatch, *nextdispatch; REQUIRE(NS_SERVER_VALID(server)); for (dispatch = ISC_LIST_HEAD(server->dispatches); dispatch != NULL; dispatch = nextdispatch) { nextdispatch = ISC_LIST_NEXT(dispatch, link); if (!all && server->dispatchgen == dispatch-> dispatchgen) continue; ISC_LIST_UNLINK(server->dispatches, dispatch, link); dns_dispatch_detach(&dispatch->dispatch); isc_mem_put(server->mctx, dispatch, sizeof(*dispatch)); } } void ns_add_reserved_dispatch(ns_server_t *server, isc_sockaddr_t *addr) { ns_dispatch_t *dispatch; in_port_t port; char addrbuf[ISC_SOCKADDR_FORMATSIZE]; isc_result_t result; unsigned int attrs, attrmask; REQUIRE(NS_SERVER_VALID(server)); port = isc_sockaddr_getport(addr); if (port == 0 || port >= 1024) return; for (dispatch = ISC_LIST_HEAD(server->dispatches); dispatch != NULL; dispatch = ISC_LIST_NEXT(dispatch, link)) { if (isc_sockaddr_equal(&dispatch->addr, addr)) break; } if (dispatch != NULL) { dispatch->dispatchgen = server->dispatchgen; return; } dispatch = isc_mem_get(server->mctx, sizeof(*dispatch)); if (dispatch == NULL) { result = ISC_R_NOMEMORY; goto cleanup; } dispatch->addr = *addr; dispatch->dispatchgen = server->dispatchgen; dispatch->dispatch = NULL; attrs = 0; attrs |= DNS_DISPATCHATTR_UDP; switch (isc_sockaddr_pf(addr)) { case AF_INET: attrs |= DNS_DISPATCHATTR_IPV4; break; case AF_INET6: attrs |= DNS_DISPATCHATTR_IPV6; break; default: result = ISC_R_NOTIMPLEMENTED; goto cleanup; } attrmask = 0; attrmask |= DNS_DISPATCHATTR_UDP; attrmask |= DNS_DISPATCHATTR_TCP; attrmask |= DNS_DISPATCHATTR_IPV4; attrmask |= DNS_DISPATCHATTR_IPV6; result = dns_dispatch_getudp(ns_g_dispatchmgr, ns_g_socketmgr, ns_g_taskmgr, &dispatch->addr, 4096, 1000, 32768, 16411, 16433, attrs, attrmask, &dispatch->dispatch); if (result != ISC_R_SUCCESS) goto cleanup; ISC_LIST_INITANDPREPEND(server->dispatches, dispatch, link); return; cleanup: if (dispatch != NULL) isc_mem_put(server->mctx, dispatch, sizeof(*dispatch)); isc_sockaddr_format(addr, addrbuf, sizeof(addrbuf)); isc_log_write(ns_g_lctx, NS_LOGCATEGORY_GENERAL, NS_LOGMODULE_SERVER, ISC_LOG_WARNING, "unable to create dispatch for reserved port %s: %s", addrbuf, isc_result_totext(result)); } static isc_result_t loadconfig(ns_server_t *server) { isc_result_t result; start_reserved_dispatches(server); result = load_configuration(ns_g_lwresdonly ? lwresd_g_conffile : ns_g_conffile, server, ISC_FALSE); if (result == ISC_R_SUCCESS) end_reserved_dispatches(server, ISC_FALSE); else isc_log_write(ns_g_lctx, NS_LOGCATEGORY_GENERAL, NS_LOGMODULE_SERVER, ISC_LOG_ERROR, "reloading configuration failed: %s", isc_result_totext(result)); return (result); } static isc_result_t reload(ns_server_t *server) { isc_result_t result; CHECK(loadconfig(server)); result = load_zones(server, ISC_FALSE); if (result != ISC_R_SUCCESS) { isc_log_write(ns_g_lctx, NS_LOGCATEGORY_GENERAL, NS_LOGMODULE_SERVER, ISC_LOG_ERROR, "reloading zones failed: %s", isc_result_totext(result)); } cleanup: return (result); } static void reconfig(ns_server_t *server) { isc_result_t result; CHECK(loadconfig(server)); result = load_new_zones(server, ISC_FALSE); if (result != ISC_R_SUCCESS) { isc_log_write(ns_g_lctx, NS_LOGCATEGORY_GENERAL, NS_LOGMODULE_SERVER, ISC_LOG_ERROR, "loading new zones failed: %s", isc_result_totext(result)); } cleanup: ; } /* * Handle a reload event (from SIGHUP). */ static void ns_server_reload(isc_task_t *task, isc_event_t *event) { ns_server_t *server = (ns_server_t *)event->ev_arg; INSIST(task = server->task); UNUSED(task); (void)reload(server); LOCK(&server->reload_event_lock); INSIST(server->reload_event == NULL); server->reload_event = event; UNLOCK(&server->reload_event_lock); } void ns_server_reloadwanted(ns_server_t *server) { LOCK(&server->reload_event_lock); if (server->reload_event != NULL) isc_task_send(server->task, &server->reload_event); UNLOCK(&server->reload_event_lock); } static char * next_token(char **stringp, const char *delim) { char *res; do { res = strsep(stringp, delim); if (res == NULL) break; } while (*res == '\0'); return (res); } /* * Find the zone specified in the control channel command 'args', * if any. If a zone is specified, point '*zonep' at it, otherwise * set '*zonep' to NULL. */ static isc_result_t zone_from_args(ns_server_t *server, char *args, dns_zone_t **zonep) { char *input, *ptr; const char *zonetxt; char *classtxt; const char *viewtxt = NULL; dns_fixedname_t name; isc_result_t result; isc_buffer_t buf; dns_view_t *view = NULL; dns_rdataclass_t rdclass; REQUIRE(zonep != NULL && *zonep == NULL); input = args; /* Skip the command name. */ ptr = next_token(&input, " \t"); if (ptr == NULL) return (ISC_R_UNEXPECTEDEND); /* Look for the zone name. */ zonetxt = next_token(&input, " \t"); if (zonetxt == NULL) return (ISC_R_SUCCESS); /* Look for the optional class name. */ classtxt = next_token(&input, " \t"); if (classtxt != NULL) { /* Look for the optional view name. */ viewtxt = next_token(&input, " \t"); } isc_buffer_init(&buf, zonetxt, strlen(zonetxt)); isc_buffer_add(&buf, strlen(zonetxt)); dns_fixedname_init(&name); result = dns_name_fromtext(dns_fixedname_name(&name), &buf, dns_rootname, ISC_FALSE, NULL); if (result != ISC_R_SUCCESS) goto fail1; if (classtxt != NULL) { isc_textregion_t r; r.base = classtxt; r.length = strlen(classtxt); result = dns_rdataclass_fromtext(&rdclass, &r); if (result != ISC_R_SUCCESS) goto fail1; } else { rdclass = dns_rdataclass_in; } if (viewtxt == NULL) viewtxt = "_default"; result = dns_viewlist_find(&server->viewlist, viewtxt, rdclass, &view); if (result != ISC_R_SUCCESS) goto fail1; result = dns_zt_find(view->zonetable, dns_fixedname_name(&name), 0, NULL, zonep); /* Partial match? */ if (result != ISC_R_SUCCESS && *zonep != NULL) dns_zone_detach(zonep); dns_view_detach(&view); fail1: return (result); } /* * Act on a "retransfer" command from the command channel. */ isc_result_t ns_server_retransfercommand(ns_server_t *server, char *args) { isc_result_t result; dns_zone_t *zone = NULL; dns_zonetype_t type; result = zone_from_args(server, args, &zone); if (result != ISC_R_SUCCESS) return (result); if (zone == NULL) return (ISC_R_UNEXPECTEDEND); type = dns_zone_gettype(zone); if (type == dns_zone_slave || type == dns_zone_stub) dns_zone_forcereload(zone); else result = ISC_R_NOTFOUND; dns_zone_detach(&zone); return (result); } /* * Act on a "reload" command from the command channel. */ isc_result_t ns_server_reloadcommand(ns_server_t *server, char *args, isc_buffer_t *text) { isc_result_t result; dns_zone_t *zone = NULL; dns_zonetype_t type; const char *msg = NULL; result = zone_from_args(server, args, &zone); if (result != ISC_R_SUCCESS) return (result); if (zone == NULL) { result = reload(server); if (result == ISC_R_SUCCESS) msg = "server reload successful"; } else { type = dns_zone_gettype(zone); if (type == dns_zone_slave || type == dns_zone_stub) { dns_zone_refresh(zone); msg = "zone refresh queued"; } else { result = dns_zone_load(zone); dns_zone_detach(&zone); switch (result) { case ISC_R_SUCCESS: msg = "zone reload successful"; break; case DNS_R_CONTINUE: msg = "zone reload queued"; result = ISC_R_SUCCESS; break; case DNS_R_UPTODATE: msg = "zone reload up-to-date"; result = ISC_R_SUCCESS; break; default: /* failure message will be generated by rndc */ break; } } } if (msg != NULL && strlen(msg) < isc_buffer_availablelength(text)) isc_buffer_putmem(text, (const unsigned char *)msg, strlen(msg) + 1); return (result); } /* * Act on a "reconfig" command from the command channel. */ isc_result_t ns_server_reconfigcommand(ns_server_t *server, char *args) { UNUSED(args); reconfig(server); return (ISC_R_SUCCESS); } /* * Act on a "refresh" command from the command channel. */ isc_result_t ns_server_refreshcommand(ns_server_t *server, char *args, isc_buffer_t *text) { isc_result_t result; dns_zone_t *zone = NULL; const unsigned char msg[] = "zone refresh queued"; result = zone_from_args(server, args, &zone); if (result != ISC_R_SUCCESS) return (result); if (zone == NULL) return (ISC_R_UNEXPECTEDEND); dns_zone_refresh(zone); dns_zone_detach(&zone); if (sizeof(msg) <= isc_buffer_availablelength(text)) isc_buffer_putmem(text, msg, sizeof(msg)); return (ISC_R_SUCCESS); } isc_result_t ns_server_togglequerylog(ns_server_t *server) { server->log_queries = server->log_queries ? ISC_FALSE : ISC_TRUE; isc_log_write(ns_g_lctx, NS_LOGCATEGORY_GENERAL, NS_LOGMODULE_SERVER, ISC_LOG_INFO, "query logging is now %s", server->log_queries ? "on" : "off"); return (ISC_R_SUCCESS); } static isc_result_t ns_listenlist_fromconfig(cfg_obj_t *listenlist, cfg_obj_t *config, ns_aclconfctx_t *actx, isc_mem_t *mctx, ns_listenlist_t **target) { isc_result_t result; cfg_listelt_t *element; ns_listenlist_t *dlist = NULL; REQUIRE(target != NULL && *target == NULL); result = ns_listenlist_create(mctx, &dlist); if (result != ISC_R_SUCCESS) return (result); for (element = cfg_list_first(listenlist); element != NULL; element = cfg_list_next(element)) { ns_listenelt_t *delt = NULL; cfg_obj_t *listener = cfg_listelt_value(element); result = ns_listenelt_fromconfig(listener, config, actx, mctx, &delt); if (result != ISC_R_SUCCESS) goto cleanup; ISC_LIST_APPEND(dlist->elts, delt, link); } *target = dlist; return (ISC_R_SUCCESS); cleanup: ns_listenlist_detach(&dlist); return (result); } /* * Create a listen list from the corresponding configuration * data structure. */ static isc_result_t ns_listenelt_fromconfig(cfg_obj_t *listener, cfg_obj_t *config, ns_aclconfctx_t *actx, isc_mem_t *mctx, ns_listenelt_t **target) { isc_result_t result; cfg_obj_t *portobj; in_port_t port; ns_listenelt_t *delt = NULL; REQUIRE(target != NULL && *target == NULL); portobj = cfg_tuple_get(listener, "port"); if (!cfg_obj_isuint32(portobj)) { if (ns_g_port != 0) { port = ns_g_port; } else { result = ns_config_getport(config, &port); if (result != ISC_R_SUCCESS) return (result); } } else { if (cfg_obj_asuint32(portobj) >= ISC_UINT16_MAX) { cfg_obj_log(portobj, ns_g_lctx, ISC_LOG_ERROR, "port value '%u' is out of range", cfg_obj_asuint32(portobj)); return (ISC_R_RANGE); } port = (in_port_t)cfg_obj_asuint32(portobj); } result = ns_listenelt_create(mctx, port, NULL, &delt); if (result != ISC_R_SUCCESS) return (result); result = ns_acl_fromconfig(cfg_tuple_get(listener, "acl"), config, actx, mctx, &delt->acl); if (result != ISC_R_SUCCESS) { ns_listenelt_destroy(delt); return (result); } *target = delt; return (ISC_R_SUCCESS); } isc_result_t ns_server_dumpstats(ns_server_t *server) { isc_result_t result; dns_zone_t *zone, *next; isc_stdtime_t now; FILE *fp = NULL; int i; int ncounters; isc_stdtime_get(&now); CHECKMF(isc_stdio_open(server->statsfile, "a", &fp), "could not open statistics dump file", server->statsfile); ncounters = DNS_STATS_NCOUNTERS; fprintf(fp, "+++ Statistics Dump +++ (%lu)\n", (unsigned long)now); for (i = 0; i < ncounters; i++) fprintf(fp, "%s %" ISC_PRINT_QUADFORMAT "u\n", dns_statscounter_names[i], server->querystats[i]); zone = NULL; for (result = dns_zone_first(server->zonemgr, &zone); result == ISC_R_SUCCESS; next = NULL, result = dns_zone_next(zone, &next), zone = next) { isc_uint64_t *zonestats = dns_zone_getstatscounters(zone); if (zonestats != NULL) { char zonename[DNS_NAME_FORMATSIZE]; dns_view_t *view; char *viewname; dns_name_format(dns_zone_getorigin(zone), zonename, sizeof(zonename)); view = dns_zone_getview(zone); viewname = view->name; for (i = 0; i < ncounters; i++) { fprintf(fp, "%s %" ISC_PRINT_QUADFORMAT "u %s", dns_statscounter_names[i], zonestats[i], zonename); if (strcmp(viewname, "_default") != 0) fprintf(fp, " %s", viewname); fprintf(fp, "\n"); } } } if (result == ISC_R_NOMORE) result = ISC_R_SUCCESS; CHECK(result); fprintf(fp, "--- Statistics Dump --- (%lu)\n", (unsigned long)now); cleanup: if (fp != NULL) (void)isc_stdio_close(fp); return (result); } static isc_result_t add_zone_tolist(dns_zone_t *zone, void *uap) { struct dumpcontext *dctx = uap; struct zonelistentry *zle; zle = isc_mem_get(dctx->mctx, sizeof *zle); if (zle == NULL) return (ISC_R_NOMEMORY); zle->zone = NULL; dns_zone_attach(zone, &zle->zone); ISC_LINK_INIT(zle, link); ISC_LIST_APPEND(ISC_LIST_TAIL(dctx->viewlist)->zonelist, zle, link); return (ISC_R_SUCCESS); } static isc_result_t add_view_tolist(struct dumpcontext *dctx, dns_view_t *view) { struct viewlistentry *vle; isc_result_t result = ISC_R_SUCCESS; vle = isc_mem_get(dctx->mctx, sizeof *vle); if (vle == NULL) return (ISC_R_NOMEMORY); vle->view = NULL; dns_view_attach(view, &vle->view); ISC_LINK_INIT(vle, link); ISC_LIST_INIT(vle->zonelist); ISC_LIST_APPEND(dctx->viewlist, vle, link); if (dctx->dumpzones) result = dns_zt_apply(view->zonetable, ISC_TRUE, add_zone_tolist, dctx); return (result); } static void dumpcontext_destroy(struct dumpcontext *dctx) { struct viewlistentry *vle; struct zonelistentry *zle; vle = ISC_LIST_HEAD(dctx->viewlist); while (vle != NULL) { ISC_LIST_UNLINK(dctx->viewlist, vle, link); zle = ISC_LIST_HEAD(vle->zonelist); while (zle != NULL) { ISC_LIST_UNLINK(vle->zonelist, zle, link); dns_zone_detach(&zle->zone); isc_mem_put(dctx->mctx, zle, sizeof *zle); zle = ISC_LIST_HEAD(vle->zonelist); } dns_view_detach(&vle->view); isc_mem_put(dctx->mctx, vle, sizeof *vle); vle = ISC_LIST_HEAD(dctx->viewlist); } if (dctx->version != NULL) dns_db_closeversion(dctx->db, &dctx->version, ISC_FALSE); if (dctx->db != NULL) dns_db_detach(&dctx->db); if (dctx->cache != NULL) dns_db_detach(&dctx->cache); if (dctx->task != NULL) isc_task_detach(&dctx->task); if (dctx->fp != NULL) (void)isc_stdio_close(dctx->fp); if (dctx->mdctx != NULL) dns_dumpctx_detach(&dctx->mdctx); isc_mem_put(dctx->mctx, dctx, sizeof *dctx); } static void dumpdone(void *arg, isc_result_t result) { struct dumpcontext *dctx = arg; char buf[1024+32]; const dns_master_style_t *style; if (result != ISC_R_SUCCESS) goto cleanup; if (dctx->mdctx != NULL) dns_dumpctx_detach(&dctx->mdctx); if (dctx->view == NULL) { dctx->view = ISC_LIST_HEAD(dctx->viewlist); if (dctx->view == NULL) goto done; INSIST(dctx->zone == NULL); } nextview: fprintf(dctx->fp, ";\n; Start view %s\n;\n", dctx->view->view->name); if (dctx->zone == NULL && dctx->cache == NULL && dctx->dumpcache) { style = &dns_master_style_cache; /* start cache dump */ if (dctx->view->view->cachedb != NULL) dns_db_attach(dctx->view->view->cachedb, &dctx->cache); if (dctx->cache != NULL) { fprintf(dctx->fp, ";\n; Cache dump of view '%s'\n;\n", dctx->view->view->name); result = dns_master_dumptostreaminc(dctx->mctx, dctx->cache, NULL, style, dctx->fp, dctx->task, dumpdone, dctx, &dctx->mdctx); if (result == DNS_R_CONTINUE) return; if (result == ISC_R_NOTIMPLEMENTED) fprintf(dctx->fp, "; %s\n", dns_result_totext(result)); else if (result != ISC_R_SUCCESS) goto cleanup; } } if (dctx->cache != NULL) { dns_adb_dump(dctx->view->view->adb, dctx->fp); dns_db_detach(&dctx->cache); } if (dctx->dumpzones) { style = &dns_master_style_full; nextzone: if (dctx->version != NULL) dns_db_closeversion(dctx->db, &dctx->version, ISC_FALSE); if (dctx->db != NULL) dns_db_detach(&dctx->db); if (dctx->zone == NULL) dctx->zone = ISC_LIST_HEAD(dctx->view->zonelist); else dctx->zone = ISC_LIST_NEXT(dctx->zone, link); if (dctx->zone != NULL) { /* start zone dump */ dns_zone_name(dctx->zone->zone, buf, sizeof(buf)); fprintf(dctx->fp, ";\n; Zone dump of '%s'\n;\n", buf); result = dns_zone_getdb(dctx->zone->zone, &dctx->db); if (result != ISC_R_SUCCESS) { fprintf(dctx->fp, "; %s\n", dns_result_totext(result)); goto nextzone; } dns_db_currentversion(dctx->db, &dctx->version); result = dns_master_dumptostreaminc(dctx->mctx, dctx->db, dctx->version, style, dctx->fp, dctx->task, dumpdone, dctx, &dctx->mdctx); if (result == DNS_R_CONTINUE) return; if (result == ISC_R_NOTIMPLEMENTED) fprintf(dctx->fp, "; %s\n", dns_result_totext(result)); if (result != ISC_R_SUCCESS) goto cleanup; } } if (dctx->view != NULL) dctx->view = ISC_LIST_NEXT(dctx->view, link); if (dctx->view != NULL) goto nextview; done: fprintf(dctx->fp, "; Dump complete\n"); result = isc_stdio_flush(dctx->fp); if (result == ISC_R_SUCCESS) isc_log_write(ns_g_lctx, NS_LOGCATEGORY_GENERAL, NS_LOGMODULE_SERVER, ISC_LOG_INFO, "dumpdb complete"); cleanup: if (result != ISC_R_SUCCESS) isc_log_write(ns_g_lctx, NS_LOGCATEGORY_GENERAL, NS_LOGMODULE_SERVER, ISC_LOG_INFO, "dumpdb failed: %s", dns_result_totext(result)); dumpcontext_destroy(dctx); } isc_result_t ns_server_dumpdb(ns_server_t *server, char *args) { struct dumpcontext *dctx = NULL; dns_view_t *view; isc_result_t result; char *ptr; const char *sep; dctx = isc_mem_get(server->mctx, sizeof(*dctx)); if (dctx == NULL) return (ISC_R_NOMEMORY); dctx->mctx = server->mctx; dctx->dumpcache = ISC_TRUE; dctx->dumpzones = ISC_FALSE; dctx->fp = NULL; ISC_LIST_INIT(dctx->viewlist); dctx->view = NULL; dctx->zone = NULL; dctx->cache = NULL; dctx->mdctx = NULL; dctx->db = NULL; dctx->cache = NULL; dctx->task = NULL; dctx->version = NULL; isc_task_attach(server->task, &dctx->task); CHECKMF(isc_stdio_open(server->dumpfile, "w", &dctx->fp), "could not open dump file", server->dumpfile); /* Skip the command name. */ ptr = next_token(&args, " \t"); if (ptr == NULL) return (ISC_R_UNEXPECTEDEND); sep = (args == NULL) ? "" : ": "; isc_log_write(ns_g_lctx, NS_LOGCATEGORY_GENERAL, NS_LOGMODULE_SERVER, ISC_LOG_INFO, "dumpdb started%s%s", sep, (args != NULL) ? args : ""); ptr = next_token(&args, " \t"); if (ptr != NULL && strcmp(ptr, "-all") == 0) { dctx->dumpzones = ISC_TRUE; dctx->dumpcache = ISC_TRUE; ptr = next_token(&args, " \t"); } else if (ptr != NULL && strcmp(ptr, "-cache") == 0) { dctx->dumpzones = ISC_FALSE; dctx->dumpcache = ISC_TRUE; ptr = next_token(&args, " \t"); } else if (ptr != NULL && strcmp(ptr, "-zones") == 0) { dctx->dumpzones = ISC_TRUE; dctx->dumpcache = ISC_FALSE; ptr = next_token(&args, " \t"); } for (view = ISC_LIST_HEAD(server->viewlist); view != NULL; view = ISC_LIST_NEXT(view, link)) { if (ptr != NULL && strcmp(view->name, ptr) != 0) continue; CHECK(add_view_tolist(dctx, view)); } dumpdone(dctx, ISC_R_SUCCESS); return (ISC_R_SUCCESS); cleanup: if (dctx != NULL) dumpcontext_destroy(dctx); return (result); } isc_result_t ns_server_dumprecursing(ns_server_t *server) { FILE *fp = NULL; isc_result_t result; CHECKMF(isc_stdio_open(server->recfile, "w", &fp), "could not open dump file", server->recfile); fprintf(fp,";\n; Recursing Queries\n;\n"); ns_interfacemgr_dumprecursing(fp, server->interfacemgr); fprintf(fp, "; Dump complete\n"); cleanup: if (fp != NULL) result = isc_stdio_close(fp); return (result); } isc_result_t ns_server_setdebuglevel(ns_server_t *server, char *args) { char *ptr; char *levelstr; char *endp; long newlevel; UNUSED(server); /* Skip the command name. */ ptr = next_token(&args, " \t"); if (ptr == NULL) return (ISC_R_UNEXPECTEDEND); /* Look for the new level name. */ levelstr = next_token(&args, " \t"); if (levelstr == NULL) { if (ns_g_debuglevel < 99) ns_g_debuglevel++; } else { newlevel = strtol(levelstr, &endp, 10); if (*endp != '\0' || newlevel < 0 || newlevel > 99) return (ISC_R_RANGE); ns_g_debuglevel = (unsigned int)newlevel; } isc_log_setdebuglevel(ns_g_lctx, ns_g_debuglevel); return (ISC_R_SUCCESS); } isc_result_t ns_server_flushcache(ns_server_t *server, char *args) { char *ptr, *viewname; dns_view_t *view; isc_boolean_t flushed = ISC_FALSE; isc_result_t result; /* Skip the command name. */ ptr = next_token(&args, " \t"); if (ptr == NULL) return (ISC_R_UNEXPECTEDEND); /* Look for the view name. */ viewname = next_token(&args, " \t"); result = isc_task_beginexclusive(server->task); RUNTIME_CHECK(result == ISC_R_SUCCESS); for (view = ISC_LIST_HEAD(server->viewlist); view != NULL; view = ISC_LIST_NEXT(view, link)) { if (viewname != NULL && strcasecmp(viewname, view->name) != 0) continue; result = dns_view_flushcache(view); if (result != ISC_R_SUCCESS) goto out; flushed = ISC_TRUE; } if (flushed) result = ISC_R_SUCCESS; else result = ISC_R_FAILURE; out: isc_task_endexclusive(server->task); return (result); } isc_result_t ns_server_flushname(ns_server_t *server, char *args) { char *ptr, *target, *viewname; dns_view_t *view; isc_boolean_t flushed = ISC_FALSE; isc_result_t result; isc_buffer_t b; dns_fixedname_t fixed; dns_name_t *name; /* Skip the command name. */ ptr = next_token(&args, " \t"); if (ptr == NULL) return (ISC_R_UNEXPECTEDEND); /* Find the domain name to flush. */ target = next_token(&args, " \t"); if (target == NULL) return (ISC_R_UNEXPECTEDEND); isc_buffer_init(&b, target, strlen(target)); isc_buffer_add(&b, strlen(target)); dns_fixedname_init(&fixed); name = dns_fixedname_name(&fixed); result = dns_name_fromtext(name, &b, dns_rootname, ISC_FALSE, NULL); if (result != ISC_R_SUCCESS) return (result); /* Look for the view name. */ viewname = next_token(&args, " \t"); result = isc_task_beginexclusive(server->task); RUNTIME_CHECK(result == ISC_R_SUCCESS); flushed = ISC_TRUE; for (view = ISC_LIST_HEAD(server->viewlist); view != NULL; view = ISC_LIST_NEXT(view, link)) { if (viewname != NULL && strcasecmp(viewname, view->name) != 0) continue; result = dns_view_flushname(view, name); if (result != ISC_R_SUCCESS) flushed = ISC_FALSE; } if (flushed) result = ISC_R_SUCCESS; else result = ISC_R_FAILURE; isc_task_endexclusive(server->task); return (result); } isc_result_t ns_server_status(ns_server_t *server, isc_buffer_t *text) { int zonecount, xferrunning, xferdeferred, soaqueries; unsigned int n; zonecount = dns_zonemgr_getcount(server->zonemgr, DNS_ZONESTATE_ANY); xferrunning = dns_zonemgr_getcount(server->zonemgr, DNS_ZONESTATE_XFERRUNNING); xferdeferred = dns_zonemgr_getcount(server->zonemgr, DNS_ZONESTATE_XFERDEFERRED); soaqueries = dns_zonemgr_getcount(server->zonemgr, DNS_ZONESTATE_SOAQUERY); n = snprintf((char *)isc_buffer_used(text), isc_buffer_availablelength(text), "number of zones: %u\n" "debug level: %d\n" "xfers running: %u\n" "xfers deferred: %u\n" "soa queries in progress: %u\n" "query logging is %s\n" "recursive clients: %d/%d\n" "tcp clients: %d/%d\n" "server is up and running", zonecount, ns_g_debuglevel, xferrunning, xferdeferred, soaqueries, server->log_queries ? "ON" : "OFF", server->recursionquota.used, server->recursionquota.max, server->tcpquota.used, server->tcpquota.max); if (n >= isc_buffer_availablelength(text)) return (ISC_R_NOSPACE); isc_buffer_add(text, n); return (ISC_R_SUCCESS); } /* * Act on a "freeze" or "unfreeze" command from the command channel. */ isc_result_t ns_server_freeze(ns_server_t *server, isc_boolean_t freeze, char *args) { isc_result_t result; dns_zone_t *zone = NULL; dns_zonetype_t type; char classstr[DNS_RDATACLASS_FORMATSIZE]; char zonename[DNS_NAME_FORMATSIZE]; dns_view_t *view; char *journal; const char *vname, *sep; isc_boolean_t frozen; result = zone_from_args(server, args, &zone); if (result != ISC_R_SUCCESS) return (result); if (zone == NULL) return (ISC_R_UNEXPECTEDEND); type = dns_zone_gettype(zone); if (type != dns_zone_master) { dns_zone_detach(&zone); return (ISC_R_NOTFOUND); } frozen = dns_zone_getupdatedisabled(zone); if (freeze) { if (frozen) result = DNS_R_FROZEN; if (result == ISC_R_SUCCESS) result = dns_zone_flush(zone); if (result == ISC_R_SUCCESS) { journal = dns_zone_getjournal(zone); if (journal != NULL) (void)isc_file_remove(journal); } } else { if (frozen) { result = dns_zone_load(zone); if (result == DNS_R_CONTINUE || result == DNS_R_UPTODATE) result = ISC_R_SUCCESS; } } if (result == ISC_R_SUCCESS) dns_zone_setupdatedisabled(zone, freeze); view = dns_zone_getview(zone); if (strcmp(view->name, "_bind") == 0 || strcmp(view->name, "_default") == 0) { vname = ""; sep = ""; } else { vname = view->name; sep = " "; } dns_rdataclass_format(dns_zone_getclass(zone), classstr, sizeof(classstr)); dns_name_format(dns_zone_getorigin(zone), zonename, sizeof(zonename)); isc_log_write(ns_g_lctx, NS_LOGCATEGORY_GENERAL, NS_LOGMODULE_SERVER, ISC_LOG_INFO, "%s zone '%s/%s'%s%s: %s", freeze ? "freezing" : "unfreezing", zonename, classstr, sep, vname, isc_result_totext(result)); dns_zone_detach(&zone); return (result); }