Issue with retry_count in socket.c and fix for the same

--

Regards,

Prakash I.

Hi,

Recently we encountered a case where clients connected to postgres server through psqlODBC driver got strucked (hanged) when server unexpectedly shutdown (hard power off or LAN cable removed). The clients were hanged forever. I analyzed the code and found a potential issue with retry count logic in function SOCK_wait_for_ready() (in socket.c).

The no_timeout variable is initialized to TRUE and is not modified if retry_count is positive number. That means if retry_count is passed then it will treat it as “no timeout”. If I understood it correctly it should work other way. If retry_count is positive number then socket should wait with timeout and not indefinitely.

I have changed the default initialization of no_timeout to FALSE. Attached is the patch. Kindly review the same and let me know comments.

Copied the function below for quick reference: socket.c: SOCK_wait_for_ready()

528 static int SOCK_wait_for_ready(SocketClass *sock, BOOL output, int retry_count)

529 {

530     int ret, gerrno;

531 #ifdef  HAVE_POLL

532     struct pollfd   fds;

533 #else

534     fd_set  fds, except_fds;

535     struct  timeval tm;

536 #endif /* HAVE_POLL */

537     BOOL    no_timeout = TRUE;

538

539     if (0 == retry_count)

540         no_timeout = FALSE;

541     else if (0  > retry_count)

542         no_timeout = TRUE;

543 #ifdef  USE_SSL

544     else if (sock->ssl == NULL)

545         no_timeout = TRUE;

546 #endif /* USE_SSL */

547     do {

548 #ifdef  HAVE_POLL

549         fds.fd = sock->socket;

550         fds.events = output ? POLLOUT : POLLIN;

551         fds.revents = 0;

552         ret = poll(&fds, 1, no_timeout ? -1 : retry_count * 1000);

553 mylog("!!!  poll ret=%d revents=%x\n", ret, fds.revents);

554 #else

555         FD_ZERO(&fds);

556         FD_ZERO(&except_fds);

557         FD_SET(sock->socket, &fds);

558         FD_SET(sock->socket, &except_fds);

559         if (!no_timeout)

560         {

561             tm.tv_sec = retry_count;

562             tm.tv_usec = 0;

563         }

564         ret = select((int) sock->socket + 1, output ? NULL : &fds, output ? &fds : NULL, &except_fds, no_timeout ? NULL : &tm);

565 #endif /* HAVE_POLL */

566         gerrno = SOCK_ERRNO;

567     } while (ret < 0 && EINTR == gerrno);

568     if (retry_count < 0)

569         retry_count *= -1;

570     if (0 == ret && retry_count > MAX_RETRY_COUNT)

571     {

572         ret = -1;

573         SOCK_set_error(sock, output ? SOCKET_WRITE_TIMEOUT : SOCKET_READ_TIMEOUT, "SOCK_wait_for_ready timeout");

574     }

575     return ret;

576 }

--

Regards,

Prakash I.