Re: PHP 4.0 Bug #5823: PHP assumes MySQL unix socket is always available

The problem I see with this patch is that it only checks to see if the unix domain socket is there, not whether it is supported on the system. It is not uncommon for the mysql client library and the server to disagree on where this unix domain socket is supposed to be. This patch would hide the error message and silently fall back to TCP and people would never know they are using the slower TCP mechanism. -Rasmus On 27 Jul 2000 dolecek@ibis.cz wrote: > From: dolecek@ibis.cz > Operating system: NetBSD-1.5_ALPHA > PHP version: 4.0.1pl2 > PHP Bug Type: MySQL related > Bug description: PHP assumes MySQL unix socket is always available > > Some pthread implementations (namely mit-pthreads, but > it might be problem also for other ones) don't support > unix sockets. Hence MySQL compiled using such threading > package does not use it. The code in php mysql > extension assumes that the unix socket is always available > on local machine (either when the host is not specified > in mysql_connect(), or it is "localhost"). This means > that it doesn't work when MySQL does not support > unix sockets. > > I've implemented simple patch solving this issue. > This is fairly heavily tested on our company servers. > The patch changes the algorithm so that > if the unix socket would be used, code first checks > if it's available and fallbacks to TCP/IP if the unix > socket does not exists. The stat() is done only once > and the value is cached for further requests, so that > there should not be any performance problems caused > with this. > > --- ext/mysql/libmysql/libmysql.c.orig Mon Mar 6 23:38:49 2000 > +++ ext/mysql/libmysql/libmysql.c Wed Jul 19 12:48:15 2000 > @@ -49,6 +49,7 @@ static my_bool mysql_client_init=0; > static MYSQL *current_mysql; > uint mysql_port=0; > my_string mysql_unix_port=0; > +static int mysql_socket_exists = -1; > > #define CLIENT_CAPABILITIES (CLIENT_LONG_PASSWORD | CLIENT_LONG_FLAG | CLIENT_LOCAL_FILES) > > @@ -1103,6 +1104,9 @@ mysql_real_connect(MYSQL *mysql,const ch > { > char buff[100],*end,*host_info; > int sock; > +#ifndef __WIN32__ > + int use_unix_socket = 0; > +#endif > ulong ip_addr; > struct sockaddr_in sock_addr; > uint pkt_length; > @@ -1165,11 +1169,37 @@ mysql_real_connect(MYSQL *mysql,const ch > */ > > #if defined(HAVE_SYS_UN_H) > - if ((!host || !strcmp(host,LOCAL_HOST)) && (unix_socket || mysql_unix_port)) > + /* > + * Check if we may use unix socket - some implementations of pthreads > + * (such as mit-pthreads) don't support unix sockets, so MySQL compiled > + * against such threading package doesn't use it even through it's configured > + * to. So we have to check whether the unix socket file actually exists > + * and use it only if it does. > + * If caller passed unix socket path explicitly, allow override this > + * check and write full error message to them later on. > + */ > + if ((!host || strcmp(host,LOCAL_HOST) == 0) > + && (unix_socket || (unix_socket = mysql_unix_port))) > + { > + if (host) { > + /* > + * Note this is thread-safe. Worst thing which would happen > + * is that stat() would have chance to be called more than > + * once here for brief period of time - not a big deal. > + */ > + if (mysql_socket_exists < 0) { > + struct stat sb; > + mysql_socket_exists = (stat(unix_socket, &sb) == 0); > + } > + > + use_unix_socket = mysql_socket_exists; > + } else > + use_unix_socket = 1; > + } > + > + if (use_unix_socket) > { > host=LOCAL_HOST; > - if (!unix_socket) > - unix_socket=mysql_unix_port; > host_info=(char*) ER(CR_LOCALHOST_CONNECTION); > DBUG_PRINT("info",("Using UNIX sock '%s'",unix_socket)); > if ((sock = socket(AF_UNIX,SOCK_STREAM,0)) == SOCKET_ERROR) > > >