UNIX / Linux keyboard.

How to enable compression with mod_gzip

Using the modified Apache web server in OpenBSD

Hosted with Apache. Hosted on OpenBSD.

Apache's httpd is a good web server. OpenBSD is a good operating system. How can we make the combination work even better?

We can add compression of the HTTP data transfer if the client's browser supports it (and unless they are using something truly ancient or bizarre, it should).

This should improve page load time and it may significantly reduce bandwidth utilization at both the server and client ends. How much of an improvement? Since images are already compressed, the improvement you get will depend on how much of a typical page is HTML (the content and the markup, including style details) as opposed to image comment.

But won't this increase the server CPU load? Surprisingly, not nearly as much as you would expect! That is, unless you are dynamically generating your pages, and maybe not even then. Linux Journal had an article about this.

If you are concerned, here is the result of using PHP to run the command:
top -d 1
on this server as it was generating this page. Really.

last pid: 48427;  load averages:  0.37,  0.43,  0.43  up 2+05:59:38    20:27:43
31 processes:  1 running, 30 sleeping
Mem: 45M Active, 308M Inact, 5884K Laundry, 178M Wired, 65M Buf, 30M Free
Swap: 1024M Total, 8616K Used, 1016M Free
10044 www          64  30    0   157M 41608K piperd   1:03   0.00% httpd
  640 root          1  20    0 18532K 10436K select   0:11   0.00% ntpd
  738 root          1  20    0 20636K  5380K select   0:03   0.00% sendmail
  656 root          1  20    0 94304K 12848K piperd   0:03   0.00% python2.7
10033 www           1  52    0 36372K  7964K piperd   0:02   0.00% php-fpm
10032 www           1  52    0 36372K  8096K accept   0:02   0.00% php-fpm
10041 www          64  52    0   141M 30040K piperd   0:02   0.00% httpd
10031 root          1  20    0 36244K  6980K kqread   0:02   0.00% php-fpm
  788 root          1  47    0 12592K  2428K nanslp   0:01   0.00% cron
10040 root          1  20    0   107M  9872K select   0:01   0.00% httpd
  554 root          1  20    0 10500K  2156K select   0:01   0.00% syslogd
  768 root          1  20    0 57812K  4844K select   0:00   0.00% sshd
  410 root          1  20    0  9560K   124K select   0:00   0.00% devd
  741 smmsp         1  20    0 20636K  4872K pause    0:00   0.00% sendmail
10043 www          64  52    0   117M 11880K piperd   0:00   0.00% httpd
10042 www          64  52    0   117M 11880K piperd   0:00   0.00% httpd
  293 root          1  40    0 10624K  1988K select   0:00   0.00% dhclient
48295 root          1  52    0 99076K 13632K select   0:00   0.00% python2.7

You can click the Reload button and you will see that things change slightly. The time (end of first line) is local time at the server, UTC-4. I would expect the CPU to be almost 100% idle, unless you happen to have loaded this page while the server is busy running a scheduled job or some other infrequent task.

What are we working with?

OpenBSD comes with something that isn't exactly Apache's httpd. The web server is the result of an OpenBSD code audit, bug patching, and hardening of something out of the Apache 1.3 web server product line. As the OpenBSD documentation says, "The OpenBSD team has added default chrooting, privilege revocation, and other security-related improvements."

Step 1 — Install the mod_gzip package

The mod_gzip Apache module is not included in the stock OpenBSD distribution. It is part of the packages and ports system. If you have downloaded the full set of packages, you have the file you need. If not, it can be as simple as this if you use csh/tcsh:

# setenv PKG_PATH ftp://ftp.openbsd.org/pub/OpenBSD/`uname -r`/packages/`uname -m`
# pkg_add mod_gzip 

Or, for bash/ksh:

# PKG_PATH=ftp://ftp.openbsd.org/pub/OpenBSD/`uname -r`/packages/`uname -m`
# pkg_add mod_gzip 

Notice that those are backquotes for command substitution! The two uname commands run first, producing output similar to 4.7 and i386. Those results are then placed into the assignment of the shell variable.

Step 2 — Modify your Apache configuration file /var/www/conf/httpd.conf

Michael Schröpl has a nice description of how to get mod_gzip to work in many situations, click here to see that. I have based what I did on his work, with some additions required for OpenBSD or any other similarly configured chroot environment.

Become root, start editing /var/www/conf/httpd.conf, and search for the string mod_gzip. When you installed that package, a post-install script within the package should have added a line loading that module. If it did not, find the area of the file where modules are loaded and add a similar line (and, to help yourself in the future, a comment!) like the following:

# Add compression
LoadModule gzip_module    /usr/lib/apache/modules/mod_gzip.so 

Now add a stanza directly below that. It should be similar to the following, but you may want to adjust the minimum and maximum file sizes:

<IfModule mod_gzip.c>
	# Turn it on:
	mod_gzip_on                     Yes
	# Enable "partial content negotiation" and the serving
	# of compressed versions if those files are available,
	# along with the updating of the compressed versions
	# when the originals are updated:
	mod_gzip_can_negotiate          Yes
	mod_gzip_static_suffix          .gz
	AddEncoding                     gzip .gz
	mod_gzip_update_static          No
	# Allow the chunks to be joined into one compresable packet
	# with a HTTP head deleted:
	mod_gzip_dechunk                Yes
	# Specify a range of file sizes in bytes for compression.
	# There's no point in compressing tiny files, and compression
	# of enormous files will delay the start of transmission.
	mod_gzip_minimum_file_size      500
	mod_gzip_maximum_file_size      6000000
	# Allow any client speaking HTTP/1.0 and later and behaving
	# in a reasonable fashion:
	mod_gzip_min_http               1000
	mod_gzip_handle_methods         GET POST
	# Compress anything named *.html and of MIME type text/*:
	mod_gzip_item_include           file  \.html$
	mod_gzip_item_include           mime  ^text/.*
	# Go ahead and compress CSS and JavaScript.  This might confuse
	# really old Netscape, but no one should be running that now.
	mod_gzip_item_include           file  \.css$
	mod_gzip_item_include           file  \.js$
	mod_gzip_item_include           mime  ^appliation/x-javascript.*
	# Do NOT compress images, there should be no point!
	mod_gzip_item_exclude           mime  ^image/.$
	# Where is the working area for temporary files and the
	# compression cache?
	mod_gzip_temp_dir               /tmp
	# See Michael Schröpl's page for optional logging in case
	# you need to do some debugging.

Step 3 — Set up the chroot environment

Here is the trick required to get this to work on OpenBSD or any other chroot environment, something that isn't frequently mentioned in the documentation.

Remember that you told your compression module to use /tmp as a working area. However, the OpenBSD web server process is using chroot to run in an unusual environment (unless you have changed /etc/rc.conf to make it behave otherwise).

The httpd process is running in a "sandbox" of sorts. Its notion of the root of the file system isn't the real root. This means that there is no ".." directory, no way to go up one level and get out of the chroot jail (or gaol as it's spelled in Britain).

What the httpd process thinks is / is really the /var/www directory. So, we have to create something that will appear to be a proper /tmp for that process. Do this:

# cd /var/www
# mkdir tmp
# chmod 1777 tmp 

Step 4 — Let's see if that works!

Stop the running web server process:

# pkill httpd 

Make sure that really worked:

# ps axuww | egrep 'PID|httpd'
# lsof -i tcp:80 

You may need to give an active request a moment or two to complete. And lsof is in /usr/local/sbin if root doesn't have that in the path.

Once the coast is clear, start a fresh web server process:

# /usr/sbin/httpd 

Yes, I probably could have done this with a HUP signal, but killing it off and starting a new one makes it a little easier to avoid sysadmin confusion. If you are concerned about not refusing a single request, look into doing it this way:

# pkill -HUP httpd 

Who cares about mod_gzip, how did you do that PHP and top trick?

This page exists as an HTML file on the server containing a block like the below. The web server process executes the PHP code on the server and inserts the output in the page.

<pre style="background: #000000; color: #00ff00;
		font-size: 85%;
		padding: 5px 3px 5px 3px;
		border: medium solid orange;">
<?php @ passthru ('top -d 1 | grep -v "^$"'); ?>

And, because of the chroot issue discussed above I had to do the following on the server to get those two commands to work:

# cd /var/www
# mkdir usr usr/bin
# cp /bin/sh bin
# cp /usr/bin/grep /usr/bin/top usr/bin
# ldd bin/sh usr/bin/*
  [ ... hmm, need to include some shared libraries ... ]
# mkdir usr/lib usr/libexec
# cp /usr/lib/libc.so.* /usr/lib/libcurses.so.* /usr/lib/libz.so.* /usr/lib
# cp /usr/libexec/ld.so usr/libexec