|
|
php.net | support | documentation | report a bug | advanced search | search howto | statistics | random bug | login |
[2000-11-16 12:15 UTC] msquillace at sogei dot it
I can't get OCI8 support to work on AIX 4.2.
During the last few weeks I have been able to configure an
Apache 1.3.12 server (with and without SSL support) to load
PHP 4 and several extensions (cURL, zlib, ClibPDF, openLDAP)
on this RS600 model F40 but the single extension that I
absolutely need (OCI8) doesn't want to work.
I use gcc and want to create a DSO module for PHP, but have
also tried static compilation and the standard AIX 4.2 cc
compiler without success; further, I started out with PHP
4.0.3 then tried 4.0.3pl1 and now downloaded the latest
CVS, to no avail.
In the following I detail the steps needed to create a
"stripped down" version of Apache+PHP4+OCI8, and describe
the behaviour I observe.
I am available for whatever further information you may
need in order to solve the problem; please reply as soon
as possible since I need to meet a deadline and, more
importantly, am trying to introduce this technology in
a company where Open Source software is still viewed
skeptically.
Prerequisite packages installation and setup
-------------------------------
01) login as root
02) gcc-2.95.2 (AIX 4.2 binaries from aixpdslib.seas.ucla.edu)
03) (GNU) make-3.79 (AIX 4.2 binaries from aixpdslib.seas.ucla.edu)
04) perl.5.005_03 (AIX 4.2 binaries from aixpdslib.seas.ucla.edu)
05) bison-1.28 [./configure, make, make install]
06) flex-2.5.4 [./configure, make, make install]
07) setup environment in order to force usage
of gcc and GNU make.
08) setup environment in order to connect to Oracle 8.1.5
and test connection via sqlplus.
09) untar in /opt/src Apache 1.3.12 and latest PHP 4 sources.
Apache+mod_so installation
--------------------
01) delete const keyword from the char *dlerror definitions
in /opt/src/apache_1.3.12/src/os/unix/os.h and
/opt/src/apache_1.3.12/src/os/unix/os-aix-dso.c
to avoid eventual make failure in the PHP4 step(s)
02) cd /opt/src/apache_1.3.12
03)./configure --enable-module=so --prefix=/opt/www
04) make
05) make install
06) uncomment ServerName directive in /opt/www/conf/httpd.conf
Apache installation test (OK)
--------------------
01) /opt/www/bin/apachectl start
02) connect to the server via netscape
03) /opt/www/bin/apachectl stop
Build PHP4 as a dynamic module via APXS with OCI8 support
(Oracle 8.1.5) and a minumum of directly supported modules.
Latest build (php4-200011151045) from snaps.php.net.
Oracle 8.1.5 is the last version supported on AIX 4.2;
I observed the exact same behaviour with Oracle 8.0.5.
------------------------------------------
01) in /ora815/lib, execute the command
ln -s libclntsh.a libclntsh.sx.8.0
to force correct detection of Oracle 8.1 in step 03.
02) cd /opt/src/php4-200011151045
mkdir /opt/www/php
mkdir /opt/www/php/bin
tar xvf ../number.tar
03) ./configure --prefix=/opt/www --mandir=/usr/man \
--with-config-file-path=/opt/www/conf \
--enable-safe-mode --with-exec-dir=/opt/www/php/bin \
--enable-track-vars --enable-magic-quotes \
--with-apxs=/opt/www/bin/apxs \
--with-mysql=no --enable-sigchild \
--enable-trans-sid --enable-versioning \
--enable-bcmath --enable-wddx --enable-xml --enable-calendar \
--without-gd --with-oci8=/ora815
04) make
During this step I only observe the following warning:
fsock.c: In function `connect_nonb':
fsock.c:182: warning: passing arg 5 of `getsockopt' from
incompatible pointer type
05) make install
06) cp php.ini-optimized /opt/www/conf/php.ini
uncomment the following in /opt/www/conf/httpd.conf:
#AddType application/x-httpd-php .php
#AddType application/x-httpd-php-source .phps
Test Apache+PHP+OCI8
------------------
01) /opt/www/bin/apachectl start
02) connect to the server via netscape ... OK
03) invoke the info.php script on the server via netscape ... OK
info.php contains:
<?php
phpinfo();
?>
04) invoke ociver1.php on the server via netscape ... NOK
ociver1.php contains:
<?php
OCIInternalDebug(1);
echo "Attempting OCILogon<br>";
error_log("Before OCILogon", 0);
$conn = OCILogon("system","manager", "ZZZ");
error_log("After OCILogon", 0);
echo "Connection handle: >".$conn."<<br>";
echo "Server Version: " . OCIServerVersion($conn);
OCILogOff($conn);
?>
This I expect to fail, since ZZZ is not a defined alias in
Oracle's tnsnames.ora; accordingly on the browser I get:
Attempting OCILogon
Warning: _oci_open_server: ORA-12154: TNS:could not resolve
service name in /opt/www/htdocs/ociver1.php on line 5
OCIDebug: _oci_open_server: FAILURE -> CLEANUP called
OCIDebug: START _oci_close_server: detaching conn=0 dbname=ZZZ
OCIDebug: _oci_close_server: closing DEAD server
OCIDebug: oci_do_connect: FAILURE -> CLEANUP called
OCIDebug: START _oci_conn_list_dtor: id=0
OCIDebug: END _oci_conn_list_dtor: id=0
Connection handle: ><
Warning: Supplied argument is not a valid OCI8-Connection resource
in /opt/www/htdocs/ociver1.php on line 8
Server Version:
Furthermore, if I tail /opt/www/logs/error_log I can correctly see:
[Thu Nov 16 17:18:09 2000] [error] Before OCILogon
[Thu Nov 16 17:18:09 2000] [error] After OCILogon
OCIDebug: START php_rshutdown_oci<br>
OCIDebug: END php_rshutdown_oci<br>
05) invoke ociver2.php on the server via netscape ... NOK
ociver2.php contains:
<?php
OCIInternalDebug(1);
echo "Attempting OCILogon<br>";
error_log("Before OCILogon", 0);
$conn = OCILogon("system","manager", "f40_7");
error_log("After OCILogon", 0);
echo "Connection handle: >".$conn."<<br>";
echo "Server Version: " . OCIServerVersion($conn);
OCILogOff($conn);
?>
In this case I expect the script to _WORK_, since f40_7 _IS_ a defined
alias in Oracle's tnsnames.ora; the browser instead sends a popup along
the lines:
The document contained no data.
Try again later, or contact the server's administrator.
The /opt/www/logs/error_log only registers:
[Thu Nov 16 17:23:17 2000] [error] Before OCILogon
and doing a ps aux | grep http I note that one of the original httpd
daemons has disappeared WITHOUT tracing anything on the
error_log (I'd expect a segfault).
Since I am using net8 (SQL*Net), I have also tried to trace it and also
activated tracing for the Oracle listener, but nothing gets written in
either trace file.
06) /opt/www/bin/apachectl stop
PatchesPull RequestsHistoryAllCommentsChangesGit/SVN commits
|
|||||||||||||||||||||||||||||||||
Copyright © 2001-2024 The PHP GroupAll rights reserved. |
Last updated: Fri Apr 19 16:01:27 2024 UTC |
I rebuilt the entire environment while writing the bug database entry, and sqlplus works in this scenario. The details of several checks follow: 1) I log in as root. 2) set | grep ORACLE returns: ORACLE_HOME=/ora815 ORACLE_SID=8T ORACLE_TERM=hft ORACLE_UID=oracle LD_PRELOAD, LD_LIBRARY_PATH, NLS_LANG and ORA_NLS33 should not be needed in AIX 4.2. 3) sqlplus system/manager@ZZZ gives ORA-12154 sqlplus system/manager@f40_7 works perfectly. 4) I now try starting apache with 3 different httpd.conf: a) user=nobody group=nobody b) user=nobody group=dba (nobody has been added to the dba group in /etc/group) c) user=apache group=dba (I created a new user apache - uid=70 - with base group=dba - gid=203 - in order to avoid having user/group ids > 60000) 5) repeating all my tests gives me the same results as in the bug report. 6) I log in as user apache. Oracle environment is the same as in 2). 7) sqlplus system/manager@ZZZ gives ORA-12154 sqlplus system/manager@f40_7 works perfectly. This morning I created a new test script, to verify PHP to Oracle connectivity using tnsping: <?php function prnit($value, $key) { echo "$value<br>"; } $t0 = microtime(); exec("tnsping f40_7", $rows); $t1 = microtime(); $ta = explode(" ", $t0); $tb = explode(" ", $t1); $t = bcsub(bcadd($tb[1], $tb[0], 6), bcadd($ta[1], $ta[0], 6), 6); echo "exec time: $t sec<br>"; array_walk($rows, 'prnit'); ?> This script works very well, producing the following output: exec time: 0.212467 sec TNS Ping Utility for IBM/AIX RISC System/6000: Version 8.1.5.0.0 - Production on 20-NOV-00 16:49:09 (c) Copyright 1997 Oracle Corporation. All rights reserved. Attempting to contact (address=(protocol=tcp)(port=1521)(host=26.2.161.96)) OK (20 msec) I believe this confirms the Oracle environment is sound, as viewed from the httpd process and, therefore from PHP.I rebuilt the entire environment while writing the bug database entry, and sqlplus works in this scenario. The details of several checks follow: 1) I log in as root. 2) set | grep ORACLE returns: ORACLE_HOME=/ora815 ORACLE_SID=8T ORACLE_TERM=hft ORACLE_UID=oracle LD_PRELOAD, LD_LIBRARY_PATH, NLS_LANG and ORA_NLS33 should not be needed in AIX 4.2. 3) sqlplus system/manager@ZZZ gives ORA-12154 sqlplus system/manager@f40_7 works perfectly. 4) I now try starting apache with 3 different httpd.conf: a) user=nobody group=nobody b) user=nobody group=dba (nobody has been added to the dba group in /etc/group) c) user=apache group=dba (I created a new user apache - uid=70 - with base group=dba - gid=203 - in order to avoid having user/group ids > 60000) 5) repeating all my tests gives me the same results as in the bug report. 6) I log in as user apache. Oracle environment is the same as in 2). 7) sqlplus system/manager@ZZZ gives ORA-12154 sqlplus system/manager@f40_7 works perfectly. This morning I created a new test script, to verify PHP to Oracle connectivity using tnsping: <?php function prnit($value, $key) { echo "$value<br>"; } $t0 = microtime(); exec("tnsping f40_7", $rows); $t1 = microtime(); $ta = explode(" ", $t0); $tb = explode(" ", $t1); $t = bcsub(bcadd($tb[1], $tb[0], 6), bcadd($ta[1], $ta[0], 6), 6); echo "exec time: $t sec<br>"; array_walk($rows, 'prnit'); ?> This script works very well, producing the following output: exec time: 0.212467 sec TNS Ping Utility for IBM/AIX RISC System/6000: Version 8.1.5.0.0 - Production on 20-NOV-00 16:49:09 (c) Copyright 1997 Oracle Corporation. All rights reserved. Attempting to contact (address=(protocol=tcp)(port=1521)(host=26.2.161.96)) OK (20 msec) I believe this confirms the Oracle environment is sound, as viewed from the httpd process and, therefore from PHP.I'm sorry to report AIX doesn't include truss nor strace. I found out a program called sctrace that you can buy from The Kernel Group for $795 (single CPU) and seems to do what truss does for Solaris and strace does for Linux, but we don't have it here. Only thing I could do was using the trace program, as per Jason's advice: trace -a -o /tmp/my_trace_log; /opt/www/bin/httpd -X; trcstop trcrpt -o /tmp/my_trace -v /tmp/my_trace_log Here follows what I believe to be the most interesting section: (about 55000 lines deleted) 107 lookuppn: /ora815/network/admin/sqlnet.ora 107 lookuppn: file not found 104 return from access. error ENOENT [73 usec] 101 access LR = D0FE3AB4 108 access /ora815/network/admin/sqlnet.ora--------- 107 lookuppn: /ora815/network/admin/ldap.ora 107 lookuppn: file not found 104 return from access. error ENOENT [45 usec] 101 access LR = D0FE3AB4 108 access /ora815/network/admin/ldap.ora--------- 107 lookuppn: /.sqlnet.ora 107 lookuppn: file not found 104 return from access. error ENOENT [20 usec] 101 access LR = D0FE3AB4 108 access /.sqlnet.ora--------- 107 lookuppn: /etc/intchg.ora 107 lookuppn: file not found 104 return from access. error ENOENT [56 usec] 101 access LR = D0FE3AB4 108 access /etc/intchg.ora--------- ([Oracle] looks for several other files, until it finds tnsnames.ora) 101 open LR = D02ACB30 107 lookuppn: /ora815/network/admin/tnsnames.ora 10D vfs number=0022, inode number=296E 15B open /ora815/network/admin/tnsnames.ora fd=4 RDONLY 104 return from open [85 usec] 101 kioctl LR = D02D9B28 14C ioctl fd=4 command=5800 arg=0000 104 return from kioctl. error ENOTTY [10 usec] 101 kfcntl LR = D02D71F0 137 fcntl /ora815/network/admin/tnsnames.ora F_SETFD return value 0000 104 return from kfcntl [5 usec] 101 kioctl LR = D02D9B28 14C ioctl fd=4 command=5800 arg=0000 104 return from kioctl. error ENOTTY [5 usec] 101 kreadv LR = D02FC0A0 163 read(4,200D6780,1000) /ora815/network/admin/tnsnames.ora 10A PFS rdwr (vp,ip)=(5FDD330,5FDD360) /ora815/network/admin/tnsnames.ora 10A PFS readi VA.S=0000.C57B bcount=0000 ip=5FDD360 104 return from kreadv [80 usec] 101 kreadv LR = D02FC0A0 163 read(4,200D6780,1000) /ora815/network/admin/tnsnames.ora 10A PFS rdwr (vp,ip)=(5FDD330,5FDD360) /ora815/network/admin/tnsnames.ora 104 return from kreadv [10 usec] 101 close LR = D02B0D34 12E close /ora815/network/admin/tnsnames.ora fd=4 104 return from close [32 usec] 101 uname LR = D0F18814 104 return from uname [11 usec] 101 getuidx LR = D02C8BC8 104 return from getuidx [2 usec] 101 kfcntl LR = D02D71F0 137 fcntl F_GETFL return value 0002 104 return from kfcntl [7 usec] 101 kfcntl LR = D02D71F0 137 fcntl F_GETFL return value 0009 104 return from kfcntl [3 usec] 101 sigprocmask LR = D02C8CD4 104 return from sigprocmask [7 usec] 101 sigprocmask LR = D02A52C0 104 return from sigprocmask [3 usec] 101 _sigaction LR = D02A5414 180 sigaction signal SIGABRT <<<<====== THIS SHOULD BE THE CULPRIT! addr new sigaction=2FF1D6D8 addr old sigaction=00000000 104 return from _sigaction [14 usec] 101 sigprocmask LR = D02A5468 104 return from sigprocmask [2 usec] 101 kfcntl LR = D02D71F0 137 fcntl F_GETFL return value 0002 104 return from kfcntl [2 usec] 101 kfcntl LR = D02D71F0 137 fcntl F_GETFL return value 0009 104 return from kfcntl [3 usec] (long list of garbage collecting calls follows) The httpd exits without segmentation faults (very quietly indeed!), but I can't see what may cause the SIGABRT.