dns-server: fix race condition issue

Dockerfile

@@ -0,0 +1,18 @@
+ FROM debian:buster-slim
+
+ RUN apt update && \
+     apt install -y git make gcc libssl-dev && \
+     git clone https://github.com/pymumu/smartdns.git --depth 1 && \
+     cd smartdns && \
+     sh ./package/build-pkg.sh --platform debian --arch `dpkg --print-architecture` && \
+     dpkg -i package/*.deb && \
+     cd / && \
+     rm -rf smartdns/ && \
+     apt autoremove -y git make gcc libssl-dev && \
+     apt clean && \
+     rm -rf /var/lib/apt/lists/*
+
+ EXPOSE 53/udp
+ VOLUME "/etc/smartdns/"
+
+ CMD ["/usr/sbin/smartdns", "-f"]

src/dns.h

@@ -110,7 +110,7 @@ struct dns_head {
 	unsigned short ancount; /* number of answer entries */
 	unsigned short nscount; /* number of authority entries */
 	unsigned short nrcount; /* number of addititional resource entries */
-} __attribute__((packed));
+} __attribute__((packed, aligned(2)));
 
 struct dns_rrs {
 	unsigned short next;

src/dns_client.c

@@ -860,7 +860,7 @@ static int _dns_client_server_add(char *server_ip, char *server_host, int port,
 		}
 
 		SSL_CTX_set_options(server_info->ssl_ctx, SSL_OP_ALL | SSL_OP_NO_SESSION_RESUMPTION_ON_RENEGOTIATION);
-
+		SSL_CTX_set_session_cache_mode(server_info->ssl_ctx, SSL_SESS_CACHE_CLIENT);
 		if (_dns_client_set_trusted_cert(server_info->ssl_ctx) != 0) {
 			tlog(TLOG_WARN, "disable check certificate for %s.", server_info->ip);
 			server_info->skip_check_cert = 1;
@@ -1513,6 +1513,11 @@ static int _DNS_client_create_socket_tcp(struct dns_server_info *server_info)
 	set_sock_keepalive(fd, 15, 3, 4);
 
 	if (connect(fd, (struct sockaddr *)&server_info->addr, server_info->ai_addrlen) != 0) {
+		if (errno == ENETUNREACH) {
+			tlog(TLOG_DEBUG, "connect %s failed, %s", server_info->ip, strerror(errno));
+			goto errout;
+		}
+
 		if (errno != EINPROGRESS) {
 			tlog(TLOG_ERROR, "connect %s failed, %s", server_info->ip, strerror(errno));
 			goto errout;
@@ -1585,6 +1590,11 @@ static int _DNS_client_create_socket_tls(struct dns_server_info *server_info, ch
 	setsockopt(fd, IPPROTO_IP, IP_TOS, &ip_tos, sizeof(ip_tos));
 
 	if (connect(fd, (struct sockaddr *)&server_info->addr, server_info->ai_addrlen) != 0) {
+		if (errno == ENETUNREACH) {
+			tlog(TLOG_DEBUG, "connect %s failed, %s", server_info->ip, strerror(errno));
+			goto errout;
+		}
+
 		if (errno != EINPROGRESS) {
 			tlog(TLOG_ERROR, "connect %s failed, %s", server_info->ip, strerror(errno));
 			goto errout;
@@ -1628,7 +1638,6 @@ errout:
 	}
 
 	if (ssl) {
-		SSL_shutdown(ssl);
 		SSL_free(ssl);
 	}
 
@@ -1729,17 +1738,19 @@ static int _dns_client_socket_ssl_send(SSL *ssl, const void *buf, int num)
 	unsigned long ssl_err = 0;
 
 	if (ssl == NULL) {
+		errno = EINVAL;
 		return -1;
 	}
 
 	ret = SSL_write(ssl, buf, num);
-	if (ret >= 0) {
+	if (ret > 0) {
 		return ret;
 	}
 
 	ssl_ret = SSL_get_error(ssl, ret);
 	switch (ssl_ret) {
 	case SSL_ERROR_NONE:
+	case SSL_ERROR_ZERO_RETURN:
 		return 0;
 		break;
 	case SSL_ERROR_WANT_READ:
@@ -1752,12 +1763,14 @@ static int _dns_client_socket_ssl_send(SSL *ssl, const void *buf, int num)
 		break;
 	case SSL_ERROR_SSL:
 		ssl_err = ERR_get_error();
-		if (ERR_GET_REASON(ssl_err) == SSL_R_UNINITIALIZED || ERR_GET_REASON(ssl_err) == SSL_R_PROTOCOL_IS_SHUTDOWN) {
+		int ssl_reason = ERR_GET_REASON(ssl_err);
+		if (ssl_reason == SSL_R_UNINITIALIZED || ssl_reason == SSL_R_PROTOCOL_IS_SHUTDOWN ||
+			ssl_reason == SSL_R_BAD_LENGTH || ssl_reason == SSL_R_SHUTDOWN_WHILE_IN_INIT) {
 			errno = EAGAIN;
 			return -1;
 		}
 
-		tlog(TLOG_DEBUG, "SSL write fail error no:  %s(%ld)\n", ERR_reason_error_string(ssl_err), ssl_err);
+		tlog(TLOG_ERROR, "SSL write fail error no:  %s(%d)\n", ERR_reason_error_string(ssl_err), ssl_reason);
 		errno = EFAULT;
 		ret = -1;
 		break;
@@ -1792,8 +1805,6 @@ static int _dns_client_socket_ssl_recv(SSL *ssl, void *buf, int num)
 	ssl_ret = SSL_get_error(ssl, ret);
 	switch (ssl_ret) {
 	case SSL_ERROR_NONE:
-		return 0;
-		break;
 	case SSL_ERROR_ZERO_RETURN:
 		return 0;
 		break;
@@ -1807,11 +1818,16 @@ static int _dns_client_socket_ssl_recv(SSL *ssl, void *buf, int num)
 		break;
 	case SSL_ERROR_SSL:
 		ssl_err = ERR_get_error();
-		if (ERR_GET_REASON(ssl_err) == SSL_R_UNINITIALIZED) {
+		int ssl_reason = ERR_GET_REASON(ssl_err);
+		if (ssl_reason == SSL_R_UNINITIALIZED) {
 			errno = EAGAIN;
 			return -1;
 		}
 
+		if (ssl_reason == SSL_R_SHUTDOWN_WHILE_IN_INIT || ssl_reason == SSL_R_PROTOCOL_IS_SHUTDOWN) {
+			return 0;
+		}
+
 		tlog(TLOG_ERROR, "SSL read fail error no: %s(%ld)\n", ERR_reason_error_string(ssl_err), ssl_err);
 		errno = EFAULT;
 		ret = -1;
@@ -2099,11 +2115,30 @@ static int _dns_client_tls_matchName(const char *host, const char *pattern, int
 	return match;
 }
 
+static int _dns_client_tls_get_cert_CN(X509 *cert, char *cn, int max_cn_len) {
+	X509_NAME *cert_name = NULL;
+
+	cert_name = X509_get_subject_name(cert);
+	if (cert_name == NULL) {
+		tlog(TLOG_ERROR, "get subject name failed.");
+		goto errout;
+	}
+
+	if (X509_NAME_get_text_by_NID(cert_name, NID_commonName, cn, max_cn_len) == -1) {
+		tlog(TLOG_ERROR, "cannot found x509 name");
+		goto errout;
+	}
+
+	return 0;
+
+errout:
+	return -1;
+}
+
 static int _dns_client_tls_verify(struct dns_server_info *server_info)
 {
 	X509 *cert = NULL;
 	X509_PUBKEY *pubkey = NULL;
-	X509_NAME *cert_name = NULL;
 	char peer_CN[256];
 	char cert_fingerprint[256];
 	int i = 0;
@@ -2127,24 +2162,20 @@ static int _dns_client_tls_verify(struct dns_server_info *server_info)
 	if (server_info->skip_check_cert == 0) {
 		long res = SSL_get_verify_result(server_info->ssl);
 		if (res != X509_V_OK) {
-			tlog(TLOG_WARN, "peer server certificate verify failed.");
+			peer_CN[0] = '\0';
+			_dns_client_tls_get_cert_CN(cert, peer_CN, sizeof(peer_CN));
+			tlog(TLOG_WARN, "peer server %s certificate verify failed", server_info->ip);
+			tlog(TLOG_WARN, "peer CN: %s", peer_CN);
 			goto errout;
 		}
 	}
 
-	cert_name = X509_get_subject_name(cert);
-	if (cert_name == NULL) {
-		tlog(TLOG_ERROR, "get subject name failed.");
-		goto errout;
-	}
-
-	if (X509_NAME_get_text_by_NID(cert_name, NID_commonName, peer_CN, 256) == -1) {
-		tlog(TLOG_ERROR, "cannot found x509 name");
+	if (_dns_client_tls_get_cert_CN(cert, peer_CN, sizeof(peer_CN)) != 0) {
+		tlog(TLOG_ERROR, "get cert CN failed.");
 		goto errout;
 	}
 
 	tlog(TLOG_DEBUG, "peer CN: %s", peer_CN);
-
 	/* check tls host */
 	tls_host_verify = _dns_client_server_get_tls_host_verify(server_info);
 	if (tls_host_verify) {
@@ -2247,10 +2278,16 @@ static int _dns_client_process_tls(struct dns_server_info *server_info, struct e
 				fd_event.events = EPOLLIN;
 			} else if (ssl_ret == SSL_ERROR_WANT_WRITE) {
 				fd_event.events = EPOLLOUT | EPOLLIN;
+			} else if (ssl_ret == SSL_ERROR_SYSCALL) {
+				if (errno != ENETUNREACH) {
+					tlog(TLOG_WARN, "Handshake with %s failed, %s", server_info->ip, strerror(errno));
+				}
+				goto errout;
 			} else {
 				unsigned long ssl_err = ERR_get_error();
-				tlog(TLOG_ERROR, "Handshake with %s failed, error no: %s(%ld)\n", server_info->ip,
-					 ERR_reason_error_string(ssl_err), ssl_err);
+				int ssl_reason = ERR_GET_REASON(ssl_err);
+				tlog(TLOG_WARN, "Handshake with %s failed, error no: %s(%d, %d, %d)\n", server_info->ip,
+					 ERR_reason_error_string(ssl_err), ret, ssl_ret, ssl_reason);
 				goto errout;
 			}
 
@@ -2289,7 +2326,7 @@ static int _dns_client_process_tls(struct dns_server_info *server_info, struct e
 
 		server_info->status = DNS_SERVER_STATUS_CONNECTED;
 		memset(&fd_event, 0, sizeof(fd_event));
-		fd_event.events = EPOLLIN;
+		fd_event.events = EPOLLIN | EPOLLOUT;
 		fd_event.data.ptr = server_info;
 		if (epoll_ctl(client.epoll_fd, EPOLL_CTL_MOD, server_info->fd, &fd_event) != 0) {
 			tlog(TLOG_ERROR, "epoll ctl failed, %s", strerror(errno));
@@ -2430,11 +2467,12 @@ static int _dns_client_send_tls(struct dns_server_info *server_info, void *packe
 	}
 
 	if (server_info->ssl == NULL) {
+		errno = EINVAL;
 		return -1;
 	}
 
 	send_len = _dns_client_socket_ssl_send(server_info->ssl, inpacket, len);
-	if (send_len < 0) {
+	if (send_len <= 0) {
 		if (errno == EAGAIN || errno == EPIPE || server_info->ssl == NULL) {
 			/* save data to buffer, and retry when EPOLLOUT is available */
 			return _dns_client_send_data_to_buffer(server_info, inpacket, len);
@@ -2480,11 +2518,12 @@ static int _dns_client_send_https(struct dns_server_info *server_info, void *pac
 	}
 
 	if (server_info->ssl == NULL) {
+		errno = EINVAL;
 		return -1;
 	}
 
 	send_len = _dns_client_socket_ssl_send(server_info->ssl, inpacket, http_len);
-	if (send_len < 0) {
+	if (send_len <= 0) {
 		if (errno == EAGAIN || errno == EPIPE || server_info->ssl == NULL) {
 			/* save data to buffer, and retry when EPOLLOUT is available */
 			return _dns_client_send_data_to_buffer(server_info, inpacket, http_len);
@@ -2507,11 +2546,12 @@ static int _dns_client_send_packet(struct dns_query_struct *query, void *packet,
 	struct dns_server_group_member *tmp = NULL;
 	int ret = 0;
 	int send_err = 0;
+	int i = 0;
 
 	query->send_tick = get_tick_count();
 
 	/* send query to all dns servers */
-	for (int i = 0; i < 2; i++) {
+	for (i = 0; i < 2; i++) {
 		pthread_mutex_lock(&client.server_list_lock);
 		list_for_each_entry_safe(group_member, tmp, &query->server_group->head, list)
 		{
@@ -2552,7 +2592,11 @@ static int _dns_client_send_packet(struct dns_query_struct *query, void *packet,
 			}
 
 			if (ret != 0) {
-				if (send_err != ENOMEM) {
+				if (send_err == ENETUNREACH) {
+					tlog(TLOG_DEBUG, "send query to %s failed, %s, type: %d", server_info->ip, strerror(send_err),
+						 server_info->type);
+					_dns_client_close_socket(server_info);
+				} else if (send_err != ENOMEM) {
 					tlog(TLOG_ERROR, "send query to %s failed, %s, type: %d", server_info->ip, strerror(send_err),
 						 server_info->type);
 				} else {

src/dns_conf.c

@@ -262,6 +262,7 @@ static int _config_server(int argc, char *argv[], dns_server_type_t type, int de
 		return -1;
 	}
 
+	ip = argv[1];
 	if (index >= DNS_MAX_SERVERS) {
 		tlog(TLOG_WARN, "exceeds max server number, %s", ip);
 		return 0;
@@ -274,8 +275,6 @@ static int _config_server(int argc, char *argv[], dns_server_type_t type, int de
 	server->httphost[0] = '\0';
 	server->tls_host_verify[0] = '\0';
 
-	ip = argv[1];
-
 	if (type == DNS_SERVER_HTTPS) {
 		if (parse_uri(ip, NULL, server->server, &port, server->path) != 0) {
 			return -1;
@@ -871,6 +870,7 @@ static int _config_bind_ip(int argc, char *argv[], DNS_BIND_TYPE type)
 		goto errout;
 	}
 
+	ip = argv[1];
 	if (index >= DNS_MAX_SERVERS) {
 		tlog(TLOG_WARN, "exceeds max server number, %s", ip);
 		return 0;
@@ -879,7 +879,6 @@ static int _config_bind_ip(int argc, char *argv[], DNS_BIND_TYPE type)
 	bind_ip = &dns_conf_bind_ip[index];
 	bind_ip->type = type;
 	bind_ip->flags = 0;
-	ip = argv[1];
 	safe_strncpy(bind_ip->ip, ip, DNS_MAX_IPLEN);
 
 	/* process extra options */

src/dns_server.c

@@ -880,6 +880,7 @@ static void _dns_server_select_possible_ipaddress(struct dns_request *request)
 
 	/* Return the most likely correct IP address */
 	/* Returns the IP with the most hits, or the last returned record is considered to be the most likely correct. */
+	pthread_mutex_lock(&request->ip_map_lock);
 	hash_for_each_safe(request->ip_map, bucket, tmp, addr_map, node)
 	{
 		if (addr_map->addr_type != request->qtype) {
@@ -896,6 +897,7 @@ static void _dns_server_select_possible_ipaddress(struct dns_request *request)
 			maxhit_addr_map = addr_map;
 		}
 	}
+	pthread_mutex_unlock(&request->ip_map_lock);
 
 	if (maxhit_addr_map && maxhit > 1) {
 		selected_addr_map = maxhit_addr_map;
@@ -966,11 +968,13 @@ static void _dns_server_request_release_complete(struct dns_request *request, in
 		_dns_server_request_complete(request);
 	}
 
+	pthread_mutex_lock(&request->ip_map_lock);
 	hash_for_each_safe(request->ip_map, bucket, tmp, addr_map, node)
 	{
 		hash_del(&addr_map->node);
 		free(addr_map);
 	}
+	pthread_mutex_unlock(&request->ip_map_lock);
 
 	_dns_server_delete_request(request);
 }
@@ -2009,7 +2013,7 @@ errout:
 	return -1;
 }
 
-static void _dns_server_prolcess_speed_check_rule(struct dns_request *request)
+static void _dns_server_process_speed_check_rule(struct dns_request *request)
 {
 	struct dns_domain_check_order *check_order = NULL;
 
@@ -2257,7 +2261,7 @@ static int _dns_server_do_query(struct dns_request *request, const char *domain,
 	}
 
 	/* process speed check rule */
-	_dns_server_prolcess_speed_check_rule(request);
+	_dns_server_process_speed_check_rule(request);
 
 	/* check and set passthrough */
 	_dns_server_check_set_passthrough(request);