resolver.c

bind-3.2.

	/*
	 * Check for any outstanding socket events.  If they exist, cancel
	 * them and let the event handlers finish the cleanup.  The resolver
	 * only needs to worry about managing the connect and send events;
	 * the dispatcher manages the recv events.
	 */
	if (RESQUERY_CONNECTING(query))
		/*
		 * Cancel the connect.
		 */
		isc_socket_cancel(query->tcpsocket, NULL,
				  ISC_SOCKCANCEL_CONNECT);
	else if (RESQUERY_SENDING(query))
		/*
		 * Cancel the pending send.
		 */
		isc_socket_cancel(dns_dispatch_getsocket(query->dispatch),
				  NULL, ISC_SOCKCANCEL_SEND);

	if (query->dispatch != NULL)
		dns_dispatch_detach(&query->dispatch);

	if (! (RESQUERY_CONNECTING(query) || RESQUERY_SENDING(query)))
		/*
		 * It's safe to destroy the query now.
		 */
		resquery_destroy(&query);
}

static void
fctx_cancelqueries(fetchctx_t *fctx, isc_boolean_t no_response) {
	resquery_t *query, *next_query;

	FCTXTRACE("cancelqueries");

	for (query = ISC_LIST_HEAD(fctx->queries);
	     query != NULL;
	     query = next_query) {
		next_query = ISC_LIST_NEXT(query, link);
		fctx_cancelquery(&query, NULL, NULL, no_response);
	}
}

static void
fctx_cleanupfinds(fetchctx_t *fctx) {
	dns_adbfind_t *find, *next_find;

	REQUIRE(ISC_LIST_EMPTY(fctx->queries));

	for (find = ISC_LIST_HEAD(fctx->finds);
	     find != NULL;
	     find = next_find) {
		next_find = ISC_LIST_NEXT(find, publink);
		ISC_LIST_UNLINK(fctx->finds, find, publink);
		dns_adb_destroyfind(&find);
	}
	fctx->find = NULL;
}

static void
fctx_cleanupforwaddrs(fetchctx_t *fctx) {
	dns_adbaddrinfo_t *addr, *next_addr;

	REQUIRE(ISC_LIST_EMPTY(fctx->queries));

	for (addr = ISC_LIST_HEAD(fctx->forwaddrs);
	     addr != NULL;
	     addr = next_addr) {
		next_addr = ISC_LIST_NEXT(addr, publink);
		ISC_LIST_UNLINK(fctx->forwaddrs, addr, publink);
		dns_adb_freeaddrinfo(fctx->adb, &addr);
	}
}

static inline void
fctx_stopeverything(fetchctx_t *fctx, isc_boolean_t no_response) {
	FCTXTRACE("stopeverything");
	fctx_cancelqueries(fctx, no_response);
	fctx_cleanupfinds(fctx);
	fctx_cleanupforwaddrs(fctx);
	fctx_stoptimer(fctx);
}

static inline void
fctx_sendevents(fetchctx_t *fctx, isc_result_t result) {
	dns_fetchevent_t *event, *next_event;
	isc_task_t *task;

	/*
	 * Caller must be holding the appropriate bucket lock.
	 */
	REQUIRE(fctx->state == fetchstate_done);

	FCTXTRACE("sendevents");

	for (event = ISC_LIST_HEAD(fctx->events);
	     event != NULL;
	     event = next_event) {
		next_event = ISC_LIST_NEXT(event, ev_link);
		ISC_LIST_UNLINK(fctx->events, event, ev_link);
		task = event->ev_sender;
		event->ev_sender = fctx;
		if (!HAVE_ANSWER(fctx))
			event->result = result;

		INSIST(result != ISC_R_SUCCESS ||
		       dns_rdataset_isassociated(event->rdataset) ||
		       fctx->type == dns_rdatatype_any ||
		       fctx->type == dns_rdatatype_sig);

		isc_task_sendanddetach(&task, (isc_event_t **)&event);
	}
}

static void
fctx_done(fetchctx_t *fctx, isc_result_t result) {
	dns_resolver_t *res;
	isc_boolean_t no_response;

	FCTXTRACE("done");

	res = fctx->res;

	if (result == ISC_R_SUCCESS)
		no_response = ISC_TRUE;
	else
		no_response = ISC_FALSE;
	fctx_stopeverything(fctx, no_response);

	LOCK(&res->buckets[fctx->bucketnum].lock);

	fctx->state = fetchstate_done;
	fctx->attributes &= ~FCTX_ATTR_ADDRWAIT;
	fctx_sendevents(fctx, result);

	UNLOCK(&res->buckets[fctx->bucketnum].lock);
}

static void
resquery_senddone(isc_task_t *task, isc_event_t *event) {
	isc_socketevent_t *sevent = (isc_socketevent_t *)event;
	resquery_t *query = event->ev_arg;

	REQUIRE(event->ev_type == ISC_SOCKEVENT_SENDDONE);

	QTRACE("senddone");

	/*
	 * XXXRTH
	 *
	 * Currently we don't wait for the senddone event before retrying
	 * a query.  This means that if we get really behind, we may end
	 * up doing extra work!
	 */

	UNUSED(task);

	INSIST(RESQUERY_SENDING(query));

	query->sends--;

	if (RESQUERY_CANCELED(query)) {
		if (query->sends == 0) {
			/*
			 * This query was canceled while the
			 * isc_socket_sendto() was in progress.
			 */
			if (query->tcpsocket != NULL)
				isc_socket_detach(&query->tcpsocket);
			resquery_destroy(&query);
		}
	} else if (sevent->result != ISC_R_SUCCESS)
		fctx_cancelquery(&query, NULL, NULL, ISC_FALSE);

	isc_event_free(&event);
}

static inline isc_result_t
fctx_addopt(dns_message_t *message) {
	dns_rdataset_t *rdataset;
	dns_rdatalist_t *rdatalist;
	dns_rdata_t *rdata;
	isc_result_t result;

	rdatalist = NULL;
	result = dns_message_gettemprdatalist(message, &rdatalist);
	if (result != ISC_R_SUCCESS)
		return (result);
	rdata = NULL;
	result = dns_message_gettemprdata(message, &rdata);
	if (result != ISC_R_SUCCESS)
		return (result);
	rdataset = NULL;
	result = dns_message_gettemprdataset(message, &rdataset);
	if (result != ISC_R_SUCCESS)
		return (result);
	dns_rdataset_init(rdataset);

	rdatalist->type = dns_rdatatype_opt;
	rdatalist->covers = 0;

	/*
	 * Set Maximum UDP buffer size.
	 */
	rdatalist->rdclass = SEND_BUFFER_SIZE;

	/*
	 * Set EXTENDED-RCODE, VERSION, and Z to 0, and the DO bit to 1.
	 */
#ifdef ISC_RFC2535
	rdatalist->ttl = DNS_MESSAGEEXTFLAG_DO;
#else
	rdatalist->ttl = 0;
#endif

	/*
	 * No EDNS options.
	 */
	rdata->data = NULL;
	rdata->length = 0;
	rdata->rdclass = rdatalist->rdclass;
	rdata->type = rdatalist->type;
	rdata->flags = 0;

	ISC_LIST_INIT(rdatalist->rdata);
	ISC_LIST_APPEND(rdatalist->rdata, rdata, link);
	dns_rdatalist_tordataset(rdatalist, rdataset);

	return (dns_message_setopt(message, rdataset));
}

static inline void
fctx_setretryinterval(fetchctx_t *fctx, unsigned int rtt) {
	unsigned int seconds;

	/*
	 * We retry every 2 seconds the first two times through the address
	 * list, and then we do exponential back-off.
	 */
	if (fctx->restarts < 3)
		seconds = 2;
	else
		seconds = (2 << (fctx->restarts - 1));

	/*
	 * Double the round-trip time and convert to seconds.
	 */
	rtt /= 500000;

	/*
	 * Always wait for at least the doubled round-trip time.
	 */
	if (seconds < rtt)
		seconds = rtt;

	/*
	 * But don't ever wait for more than 30 seconds.
	 */
	if (seconds > 30)
		seconds = 30;

	isc_interval_set(&fctx->interval, seconds, 0);
}

static isc_result_t
fctx_query(fetchctx_t *fctx, dns_adbaddrinfo_t *addrinfo,
	   unsigned int options)
{
	dns_resolver_t *res;
	isc_task_t *task;
	isc_result_t result;
	resquery_t *query;

	FCTXTRACE("query");

	res = fctx->res;
	task = res->buckets[fctx->bucketnum].task;

	fctx_setretryinterval(fctx, addrinfo->srtt);
	result = fctx_startidletimer(fctx);
	if (result != ISC_R_SUCCESS)
		return (result);

	dns_message_reset(fctx->rmessage, DNS_MESSAGE_INTENTPARSE);

	query = isc_mem_get(res->mctx, sizeof *query);
	if (query == NULL) {
		result = ISC_R_NOMEMORY;
		goto stop_idle_timer;
	}
	query->mctx = res->mctx;
	query->options = options;
	query->attributes = 0;
	query->sends = 0;
	query->connects = 0;
	/*
	 * Note that the caller MUST guarantee that 'addrinfo' will remain
	 * valid until this query is canceled.
	 */
	query->addrinfo = addrinfo;
	result = isc_time_now(&query->start);
	if (result != ISC_R_SUCCESS)
		goto cleanup_query;

	/*
	 * If this is a TCP query, then we need to make a socket and
	 * a dispatch for it here.  Otherwise we use the resolver's
	 * shared dispatch.
	 */
	query->dispatchmgr = res->dispatchmgr;
	query->dispatch = NULL;
	query->tcpsocket = NULL;
	if ((query->options & DNS_FETCHOPT_TCP) != 0) {
		isc_sockaddr_t addr;
		int pf;

		pf = isc_sockaddr_pf(&addrinfo->sockaddr);

		switch (pf) {
		case PF_INET:
			result = dns_dispatch_getlocaladdress(res->dispatchv4,
							      &addr);
			break;
		case PF_INET6:
			result = dns_dispatch_getlocaladdress(res->dispatchv6,
							      &addr);
			break;
		default:
			result = ISC_R_NOTIMPLEMENTED;
			break;
		}
		if (result != ISC_R_SUCCESS)
			goto cleanup_query;

		isc_sockaddr_setport(&addr, 0);

		result = isc_socket_create(res->socketmgr, pf,
					   isc_sockettype_tcp,
					   &query->tcpsocket);
		if (result != ISC_R_SUCCESS)
			goto cleanup_query;

		result = isc_socket_bind(query->tcpsocket, &addr);
		if (result != ISC_R_SUCCESS)
			goto cleanup_socket;

		/*
		 * A dispatch will be created once the connect succeeds.
		 */
	} else {
		switch (isc_sockaddr_pf(&addrinfo->sockaddr)) {
		case PF_INET:
			dns_dispatch_attach(res->dispatchv4, &query->dispatch);
			break;
		case PF_INET6:
			dns_dispatch_attach(res->dispatchv6, &query->dispatch);
			break;
		default:
			result = ISC_R_NOTIMPLEMENTED;
			goto cleanup_query;
		}
		/*
		 * We should always have a valid dispatcher here.  If we
		 * don't support a protocol family, then its dispatcher
		 * will be NULL, but we shouldn't be finding addresses for
		 * protocol types we don't support, so the dispatcher
		 * we found should never be NULL.
		 */
		INSIST(query->dispatch != NULL);
	}

	query->dispentry = NULL;
	query->fctx = fctx;
	query->tsig = NULL;
	query->tsigkey = NULL;
	ISC_LINK_INIT(query, link);
	query->magic = QUERY_MAGIC;

	if ((query->options & DNS_FETCHOPT_TCP) != 0) {
		/*
		 * Connect to the remote server.
		 *
		 * XXXRTH  Should we attach to the socket?
		 */
		result = isc_socket_connect(query->tcpsocket,
					    &addrinfo->sockaddr, task,
					    resquery_connected, query);
		if (result != ISC_R_SUCCESS)
			goto cleanup_socket;
		query->connects++;
		QTRACE("connecting via TCP");
	} else {
		result = resquery_send(query);
		if (result != ISC_R_SUCCESS)
			goto cleanup_dispatch;
	}

	ISC_LIST_APPEND(fctx->queries, query, link);

	return (ISC_R_SUCCESS);

 cleanup_socket:
	isc_socket_detach(&query->tcpsocket);

 cleanup_dispatch:
	if (query->dispatch != NULL)
		dns_dispatch_detach(&query->dispatch);

 cleanup_query:
	query->magic = 0;
	isc_mem_put(res->mctx, query, sizeof *query);

 stop_idle_timer:
	fctx_stopidletimer(fctx);

	return (result);
}

static isc_result_t
resquery_send(resquery_t *query) {
	fetchctx_t *fctx;
	isc_result_t result;
	dns_name_t *qname = NULL;
	dns_rdataset_t *qrdataset = NULL;
	isc_region_t r;
	dns_resolver_t *res;
	isc_task_t *task;
	isc_socket_t *socket;
	isc_buffer_t tcpbuffer;
	isc_sockaddr_t *address;
	isc_buffer_t *buffer;
	isc_netaddr_t ipaddr;
	dns_tsigkey_t *tsigkey = NULL;
	dns_peer_t *peer = NULL;
	isc_boolean_t useedns;
	dns_compress_t cctx;
	isc_boolean_t cleanup_cctx = ISC_FALSE;

	fctx = query->fctx;
	QTRACE("send");

	res = fctx->res;
	task = res->buckets[fctx->bucketnum].task;
	address = NULL;

	if ((query->options & DNS_FETCHOPT_TCP) != 0) {
		/*
		 * Reserve space for the TCP message length.
		 */
		isc_buffer_init(&tcpbuffer, query->data, sizeof(query->data));
		isc_buffer_init(&query->buffer, query->data + 2,
				sizeof(query->data) - 2);
		buffer = &tcpbuffer;
	} else {
		isc_buffer_init(&query->buffer, query->data,
				sizeof(query->data));
		buffer = &query->buffer;
	}

	result = dns_message_gettempname(fctx->qmessage, &qname);
	if (result != ISC_R_SUCCESS)
		goto cleanup_temps;
	result = dns_message_gettemprdataset(fctx->qmessage, &qrdataset);
	if (result != ISC_R_SUCCESS)
		goto cleanup_temps;

	/*
	 * Get a query id from the dispatch.
	 */
	result = dns_dispatch_addresponse(query->dispatch,
					  &query->addrinfo->sockaddr,
					  task,
					  resquery_response,
					  query,
					  &query->id,
					  &query->dispentry);
	if (result != ISC_R_SUCCESS)
		goto cleanup_temps;

	fctx->qmessage->opcode = dns_opcode_query;

	/*
	 * Set up question.
	 */
	dns_name_init(qname, NULL);
	dns_name_clone(&fctx->name, qname);
	dns_rdataset_init(qrdataset);
	dns_rdataset_makequestion(qrdataset, res->rdclass, fctx->type);
	ISC_LIST_APPEND(qname->list, qrdataset, link);
	dns_message_addname(fctx->qmessage, qname, DNS_SECTION_QUESTION);
	qname = NULL;
	qrdataset = NULL;

	/*
	 * Set RD if the client has requested that we do a recursive query,
	 * or if we're sending to a forwarder.
	 */
	if ((query->options & DNS_FETCHOPT_RECURSIVE) != 0 ||
	    ISFORWARDER(query->addrinfo))
		fctx->qmessage->flags |= DNS_MESSAGEFLAG_RD;

	/*
	 * We don't have to set opcode because it defaults to query.
	 */
	fctx->qmessage->id = query->id;

	/*
	 * Convert the question to wire format.
	 */
	result = dns_compress_init(&cctx, -1, fctx->res->mctx);
	if (result != ISC_R_SUCCESS)
		goto cleanup_message;
	cleanup_cctx = ISC_TRUE;