Evaggelos Balaskas - System Engineer

Yesterday evening i had the pleasure to watch my apache crashing till the entire memory of my vps server was been consumed.
I had the opportunity to see a memory leak and drink a couple of beers among good friends.
Friends that can support you (psychological) till you find the bug (is it?) and fix it.

So lets begin our journey:

My blog engine (flatpress) has a identi.ca/twitter plugin for posting entries on my blog.
I’ve connected it with my identi.ca account and i ‘ve done a little hack to add a microblogging category to separate my rss feed from my blogging rss feed (category=1)

So the main problem was(is) that the identica.png image doesnt get the correct file path from php variable.
It should be something like that:

blog/fp-plugins/identicaconnect/res/identica.png

but it seems to be:

https://balaskas.gr/blog/https://balaskas.gr/blog/blog/fp-plugins/identicaconnect/res/identica.png

That would be easy to fix, right?
That was what i thought too.

But in the process or fixing it, i saw the below error on my apache logs:

“PHP Notice: Undefined index: PATH_INFO”

I fired up my php.info page and saw that there wasnt any value for the $_SERVER[’PATH_INFO’]
In fact there wasnt any $_SERVER[’PATH_INFO’] in PHP Variables !!!

WTF ?

I was searching for an answer on google and i was noticing that my site was inaccessible.

pgrep httpd | wc -l

showed me about 200 apache threads and rising really fast.

dmesg complaint about resource and at that moment my vps crashed for the first time with a memory leak in console !!!

My previous apache installation was : httpd 2.0.64 + php-5.3.3 + suhosin-patch-5.3.3-0.9.10.patch + mod_evasive + eaccelerator-0.9.6.1 and my apache custom compilation options were:

./configure     
        --enable-dav 
        --enable-rewrite 
        --enable-ssl 
        --enable-so 
        --enable-proxy 
        --enable-headers 
        --enable-deflate 
        --enable-cache 
        --enable-disk-cache

my php compilation options were:

./configure  
        --with-zlib     
        --with-openssl  
        --with-gd       
        --enable-mbstring 
        --with-apxs2=/usr/local/apache2/bin/apxs 
        --with-mysql 
        --with-mcrypt 
        --with-curl

When i saw the memory leak, my first (and only) thought was: killapache.pl !

In a heartbeat, i was compiling httpd-2.2.20 + php-5.3.8 + suhosin-patch-5.3.7-0.9.10.patch + eaccelerator-0.9.6.1 + mod_evasive, i had moved my /usr/local/apache2 folder to apache2.bak and installed the newest (and hopefully most secure) version of apache & php.

I have pretty well document all of my installations process and i am keeping comments for every line in configuration files i have ever changed. So to setup up httpd 2.2.20 was in indeed a matter of minutes.

I was feeling lucky and confident.

I started apache and fired up my blog.
I was tailing error logs too.

BUM !!!!

apache had just crashed again !!!!

WTF^2 ?

How can a null php variable, crash apache with a memory leak and open about a million threads?
After debugging it, i fix it by just putting an isset function in front of $_SERVER[’PATH_INFO’] php variable !!!!

Too much trouble to fix (i didnt) the path of an image in my blog.

So my question is this:

• Is this an apache bug ?
• Is this a php bug ? or
• Is it a software bud (flatpress) ?