Be more wary about OpenSSL not setting errno on error.

tglsfdc · tglsfdc · commit ebbd499d4b55 · 2023-12-11T11:51:56.000-05:00
OpenSSL will sometimes return SSL_ERROR_SYSCALL without having set errno; this is apparently a reflection of recv(2)'s habit of not setting errno when reporting EOF. Ensure that we treat such cases the same as read EOF. Previously, we'd frequently report them like "could not accept SSL connection: Success" which is confusing, or worse report them with an unrelated errno left over from some previous syscall. To fix, ensure that errno is zeroed immediately before the call, and report its value only when it's not zero afterwards; otherwise report EOF. For consistency, I've applied the same coding pattern in libpq's pqsecure_raw_read(). Bare recv(2) shouldn't really return -1 without setting errno, but in case it does we might as well cope. Per report from Andres Freund. Back-patch to all supported versions. Discussion: https://postgr.es/m/20231208181451.deqnflwxqoehhxpe@awork3.anarazel.de
diff --git a/src/backend/libpq/be-secure-openssl.c b/src/backend/libpq/be-secure-openssl.c
@@ -460,6 +460,7 @@ be_tls_open_server(Port *port)
 	 * per-thread error queue following another call to an OpenSSL I/O
 	 * routine.
 	 */
+	errno = 0;
 	ERR_clear_error();
 	r = SSL_accept(port->ssl);
 	if (r <= 0)
@@ -496,7 +497,7 @@ be_tls_open_server(Port *port)
 										 WAIT_EVENT_SSL_OPEN_SERVER);
 				goto aloop;
 			case SSL_ERROR_SYSCALL:
-				if (r < 0)
+				if (r < 0 && errno != 0)
 					ereport(COMMERROR,
 							(errcode_for_socket_access(),
 							 errmsg("could not accept SSL connection: %m")));
@@ -732,7 +733,7 @@ be_tls_read(Port *port, void *ptr, size_t len, int *waitfor)
 			break;
 		case SSL_ERROR_SYSCALL:
 			/* leave it to caller to ereport the value of errno */
-			if (n != -1)
+			if (n != -1 || errno == 0)
 			{
 				errno = ECONNRESET;
 				n = -1;
@@ -790,8 +791,14 @@ be_tls_write(Port *port, void *ptr, size_t len, int *waitfor)
 			n = -1;
 			break;
 		case SSL_ERROR_SYSCALL:
-			/* leave it to caller to ereport the value of errno */
-			if (n != -1)
+
+			/*
+			 * Leave it to caller to ereport the value of errno.  However, if
+			 * errno is still zero then assume it's a read EOF situation, and
+			 * report ECONNRESET.  (This seems possible because SSL_write can
+			 * also do reads.)
+			 */
+			if (n != -1 || errno == 0)
 			{
 				errno = ECONNRESET;
 				n = -1;
diff --git a/src/backend/libpq/pqcomm.c b/src/backend/libpq/pqcomm.c
@@ -936,6 +936,8 @@ pq_recvbuf(void)
 	{
 		int			r;
 
+		errno = 0;
+
 		r = secure_read(MyProcPort, PqRecvBuffer + PqRecvLength,
 						PQ_RECV_BUFFER_SIZE - PqRecvLength);
 
@@ -948,10 +950,13 @@ pq_recvbuf(void)
 			 * Careful: an ereport() that tries to write to the client would
 			 * cause recursion to here, leading to stack overflow and core
 			 * dump!  This message must go *only* to the postmaster log.
+			 *
+			 * If errno is zero, assume it's EOF and let the caller complain.
 			 */
-			ereport(COMMERROR,
-					(errcode_for_socket_access(),
-					 errmsg("could not receive data from client: %m")));
+			if (errno != 0)
+				ereport(COMMERROR,
+						(errcode_for_socket_access(),
+						 errmsg("could not receive data from client: %m")));
 			return EOF;
 		}
 		if (r == 0)
@@ -1028,6 +1033,8 @@ pq_getbyte_if_available(unsigned char *c)
 	/* Put the socket into non-blocking mode */
 	socket_set_nonblocking(true);
 
+	errno = 0;
+
 	r = secure_read(MyProcPort, c, 1);
 	if (r < 0)
 	{
@@ -1044,10 +1051,13 @@ pq_getbyte_if_available(unsigned char *c)
 			 * Careful: an ereport() that tries to write to the client would
 			 * cause recursion to here, leading to stack overflow and core
 			 * dump!  This message must go *only* to the postmaster log.
+			 *
+			 * If errno is zero, assume it's EOF and let the caller complain.
 			 */
-			ereport(COMMERROR,
-					(errcode_for_socket_access(),
-					 errmsg("could not receive data from client: %m")));
+			if (errno != 0)
+				ereport(COMMERROR,
+						(errcode_for_socket_access(),
+						 errmsg("could not receive data from client: %m")));
 			r = EOF;
 		}
 	}
diff --git a/src/interfaces/libpq/fe-secure-openssl.c b/src/interfaces/libpq/fe-secure-openssl.c
@@ -207,7 +207,7 @@ pgtls_read(PGconn *conn, void *ptr, size_t len)
 			 */
 			goto rloop;
 		case SSL_ERROR_SYSCALL:
-			if (n < 0)
+			if (n < 0 && SOCK_ERRNO != 0)
 			{
 				result_errno = SOCK_ERRNO;
 				if (result_errno == EPIPE ||
@@ -308,7 +308,13 @@ pgtls_write(PGconn *conn, const void *ptr, size_t len)
 			n = 0;
 			break;
 		case SSL_ERROR_SYSCALL:
-			if (n < 0)
+
+			/*
+			 * If errno is still zero then assume it's a read EOF situation,
+			 * and report EOF.  (This seems possible because SSL_write can
+			 * also do reads.)
+			 */
+			if (n < 0 && SOCK_ERRNO != 0)
 			{
 				result_errno = SOCK_ERRNO;
 				if (result_errno == EPIPE || result_errno == ECONNRESET)
@@ -1523,11 +1529,12 @@ open_client_SSL(PGconn *conn)
 					 * was using the system CA pool. For other errors, log
 					 * them using the normal SYSCALL logging.
 					 */
-					if (!save_errno && vcode == X509_V_ERR_UNABLE_TO_GET_ISSUER_CERT_LOCALLY &&
+					if (save_errno == 0 &&
+						vcode == X509_V_ERR_UNABLE_TO_GET_ISSUER_CERT_LOCALLY &&
 						strcmp(conn->sslrootcert, "system") == 0)
 						libpq_append_conn_error(conn, "SSL error: certificate verify failed: %s",
 												X509_verify_cert_error_string(vcode));
-					else if (r == -1)
+					else if (r == -1 && save_errno != 0)
 						libpq_append_conn_error(conn, "SSL SYSCALL error: %s",
 												SOCK_STRERROR(save_errno, sebuf, sizeof(sebuf)));
 					else
diff --git a/src/interfaces/libpq/fe-secure.c b/src/interfaces/libpq/fe-secure.c
@@ -233,6 +233,8 @@ pqsecure_raw_read(PGconn *conn, void *ptr, size_t len)
 	int			result_errno = 0;
 	char		sebuf[PG_STRERROR_R_BUFLEN];
 
+	SOCK_ERRNO_SET(0);
+
 	n = recv(conn->sock, ptr, len, 0);
 
 	if (n < 0)
@@ -259,6 +261,11 @@ pqsecure_raw_read(PGconn *conn, void *ptr, size_t len)
 										"\tbefore or while processing the request.");
 				break;
 
+			case 0:
+				/* If errno didn't get set, treat it as regular EOF */
+				n = 0;
+				break;
+
 			default:
 				libpq_append_conn_error(conn, "could not receive data from server: %s",
 										SOCK_STRERROR(result_errno,
