pig-monkey.com - vps

An Ubuntu VPS for Django

2011-07-19T00:00:00-07:00

Three years ago I wrote a guide to building a VPS web server for serving sites in a PHP environment. That setup served me well for some time, but most of the sites I run now – including this one – are now written in Python. Earlier this year I built another web server to reflect this. It’s similar to before; I still use Ubuntu and I still like to serve pages with nginx. But PHP has been replaced with Python, and many of the packages used to build the environment have changed as a result. As with the last time, I decided to compile my notes into a guide, both for my own reference and in case anyone else would like to duplicate it. So far, the server has proven to be fast and efficient. It serves Python using uWSGI, uses a PostgreSQL database, and includes a simple mail server provided by Postfix. I think it’s a good setup for serving simple Django-based websites.

Basic Setup

As with last time, I recommend following Slicehost’s basic server setup article. It discusses user administration, SSH security, and firewalls. I no longer use Slicehost as my VPS provider, but I find that Slicehost’s articles provide an excellent base regardless of the host.

Packages

Packages should upgraded immediately to address any known security vulnerabilities.

$ sudo apt-get update
$ sudo apt-get upgrade

After the repositories have been updated, I install some essential packages.

1	`$ sudo apt-get install build-essential screen dnsutils`

Build-essential includes necessary tools to compile programs. I am incapable of using a computer that does not have screen on it, so that gets installed too. The third package, dnsutils, is optional, but includes dig which is useful for troubleshooting DNS issues.

DenyHosts

Slicehost’s setup article recommends turning off password authentication in SSH, forcing users to login with keys only. I use keys whenever I can, but I appreciate the option of being able to login to my server from any computer, when I may or may not have my SSH key with me. So I leave password authentication enabled. This presents the possibility of brute-force attacks. Enter DenyHosts. DenyHosts, which I have discussed previously attempts to protect against SSH attacks by banning hosts after a certain number of failed login attempts. When password authentication is enabled, running DenyHosts is a smart move.

$ sudo apt-get install denyhosts
$ sudo vim /etc/denyhosts.conf

Personalize the Environment

I use update-alternatives to set my default editor to vim.

1	`$ sudo update-alternatives --config editor`

All of my personal configuration files are kept in a github repository. I’ll check out that repository into ~/src and install the files.

$ mkdir ~/src
$ cd ~/src
$ sudo apt-get install git-core
$ git clone git://github.com/pigmonkey/dotfiles.git
$ ln -s ~/src/dotfiles/bash_profile ~/.bash_profile
$ ln -s ~/src/dotfiles/bashrc ~/.bashrc
$ ln -s ~/src/dotfiles/bash_aliases ~/.bash_aliases
$ ln -s ~/src/dotfiles/bash_colors ~/.bash_colors
$ ln -s ~/src/dotfiles/vimrc ~/.vimrc
$ ln -s ~/src/dotfiles/vim ~/.vim
$ ln -s ~/src/dotfiles/screenrc ~/.screenrc
$ source ~/.bash_profile

Time

The next step is to set the server’s timezone and sync the clock with NTP.

$ sudo dpkg-reconfigure tzdata
$ sudo apt-get install ntp

Rootkits

There’s no reason not to run both chkrootkit and rkhunter to check for rootkits and related vulnerabilities.

chrkrootkit

Slicehost has an excellent article for setting up and using chkrootkit.

Later on I’ll be installing some Python development packages. One of them creates a directory called /usr/lib/pymodules/python2.6/.path, which sets off a warning in chkrootkit. Part of chkrootkit’s desgin philosophy is to not include any whitelists: if chkrootkit finds something that it doesn’t like, you’re going to hear about it. I have cron run chkrootkit nightly and I want to receive any warnings, but I don’t want to receive the same false positive every morning in my inbox.

The solution is to create a file that contains chkrootkit’s warning. I call that file whitelist and store it in the same directory as chkrootkit. When chkrootkit is run, any output is redirected to a file. That file is compared to the whitelist using diff and the output of that – if any – is then read. At the end, the file containing chkrootkit’s output is deleted so that the working directory is ready for the next run. The effect is that I only hear warnings from chkrootkit that I have not explicit whitelisted. All of this can be accomplished in a single crontab entry.

0 3 * * * (cd /home/demo/src/chkrootkit-0.49; ./chkrootkit -q > message 2>&1; diff -w whitelist message; rm -f message)

rkhunter

I’m sure it doesn’t surprise you that I’m going to recommend reading Slicehost’s article on rkhunter.

Unlike chkrootkit, rkhunter does allow for whitelists. On a clean Ubuntu 10.04 system, I find that I need to whitelist a few items in the config.

$ sudo vim /etc/rkhunter.conf

SCRIPTWHITELIST="/usr/sbin/adduser /usr/bin/ldd /bin/which"
ALLOWHIDDENDIR="/dev/.udev /dev/.initramfs"
APP_WHITELIST="openssl:0.9.8k gpg:1.4.10"

$ sudo /usr/local/bin/rkhunter --propupd

The script that my cronjob runs is slightly different from the one demonstrated in the Slicehost article. Their script executes a few commands, groups the output together, and sends it to mail to email the system administrator. This results in daily emails, regardless of whether rkhunter finds any warnings or not. My script is simpler and does not result in so many messages.

#!/bin/sh
/usr/local/bin/rkhunter --versioncheck -q
/usr/local/bin/rkhunter --update -q
/usr/local/bin/rkhunter --cronjob --report-warnings-only

The version check and update commands both have the -q switch, which disables any output – I don’t care to know whether rkhunter updated itself or not. The final line actually executes the scan. Notice that there’s no reference to mail. This script does not send any messages. The reason for that is that rkhunter itself provides the mail functionality. Inside of /etc/rkhunter.conf there is a MAIL-ON-WARNING variable. As long as the machine has an smtp server on it (which I’ll get to later in this guide), simply filling in this variable will result in any warnings being emailed to the system administrator.

Web Server

With the basics complete, it’s time to start serving something! In my previous article I covered serving a PHP-based Wordpress site via FastCGI and nginx. This time around the stack will be different: nginx, uWSGI, Python, and Django.

A few basic packages will help flesh out the server’s Python development environment:

$ sudo apt-get install python-psycopg2 python-setuptools python2.6-dev psmisc python-imaging locate python-dateutil libxml2-dev python-software-properties

uWSGI

Installing uWSGI is a simple matter of compiling it and moving the resulting binary into one of your system’s bin directories.

$ cd ~/src/
$ wget http://projects.unbit.it/downloads/uwsgi-0.9.8.tar.gz
$ tar xvzf ~/uwsgi-0.9.8.tar.gz
$ cd uwsgi-0.9.8/
$ make -f Makefile.Py26
$ sudo cp uwsgi /usr/local/sbin

nginx

The nginx package in Ubuntu’s official repositories is always notoriously outdated. It used to be you had to compile the server from source, but there is now an Ubuntu PPA for the latest stable versions. As described by the nginx wiki, all that is needed is to add the PPA to your sources.list and apt-get away!

$ sudo add-apt-repository ppa:nginx/stable
$ sudo apt-get update 
$ sudo apt-get install nginx

Python and Django

If you do Python development and haven’t heard of virtualenv, it is well worth reading up on. It allows the user to create an isolated, virtual Python environment for each project. This helps immensely when developing (or serving) multiple projects on a single machine. Needless to say, I consider it to be a required package.

Install

I’ll be installing virtualenv and virtualenvwrapper (a set of scripts to facilitate working with virtual environments). I also prefer pip over easy_install for managing Python packages.

$ sudo easy_install -U pip 
$ sudo pip install virtualenv
$ sudo pip install virtualenvwrapper

Setup a virtual environment

Virtual environments can be stored wherever you fancy. For now, I keep them in a hidden folder in my home directory. For these examples, I’ll setup an environment called myproject.

$ mkdir ~/.virtualenvs
$ cd ~/.virtualenvs
$ virtualenv --no-site-packages --distribute myproject

Notice the --no-site-packages switch. That tells virtualenv to create this environment without any of the Python packages already installed, creating a completely fresh, clean environment. The --distribute switch causes the new virtual environment to be setup with distribute, a replacement for the old and rather broken setuptools.

All that’s needed to get virtualenvwrapper up and running is to add two lines to your .bashrc and re-source the file.

$ vim ~/.bashrc

export WORKON_HOME=$HOME/.virtualenvs
source /usr/local/bin/virtualenvwrapper.sh

$ . ~/.bashrc

We can now use commands like workon to ease the process of activating a certain environment.

I’ll go ahead and install yolk in the environment to help manage packages.

$ workon myproject
$ pip install yolk
$ yolk -l

The last command will cause yolk to list all packages installed in the environment. Try deactivating the environment and then running yolk again.

$ deactivate
$ yolk -l
yolk: command not found

‘yolk’ wasn’t found, because it was only installed within the virtual environment. Neat!

Install Django

Finally, it’s time to install Django! The process is simple enough.

$ workon myproject
$ pip install django

And that’s it!

The Python Imaging Library is likely to be needed for any Django project. I installed it in the beginning of this section, but because I used the --no-site-packages when creating my virtual environment, it is not available for use within the project. To fix that, I’ll just link the package in. I also previously installed psyopg2, which Python will need to communicate with my PostgreSQL database, so I’ll link that in as well. psyopg2 depends on mx, which was also previously installed but still must be made available in the environment.

$ cdsitepackages
$ ln -s /usr/lib/python2.6/dist-packages/PIL
$ ln -s /usr/lib/python2.6/dist-packages/PIL.pth
$ ln -s /usr/lib/python2.6/dist-packages/psycopg2
$ ln -s /usr/lib/python2.6/dist-packages/mx

That wasn’t too painful!

Create a Django project

While I’m still in the virtual environment, I’ll go ahead and create a new Django project. The project will have the same name as the environment: myproject. For this tutorial, I’ll stick with the precedence set by the Slicehost tutorials and use demo as the name of both my user and group on the server.

I like to keep my sites in the /srv/ directory. I structure them so that the code that runs the site, any public files, logs, and backups are all stored in separate sub-directories.

$ cd /srv/
$ sudo mkdir -p myproject.com/{code,public,logs,backup}
$ sudo mkdir -p myproject.com/public/{media,static}
$ sudo chown -R demo:demo myproject.com
$ cd myproject.com
$ sudo chown -R :www-data logs public
$ sudo chmod -R g+w logs public
$ cd code/
$ django-admin.py startproject myproject

Notice that the logs and public directories were chowned to the www-data group. That is the name of the user and group that nginx will run as. The web server will need permissions to write to those locations.

Save Requirements

With the environment setup and all the necessary packages installed, now is a good time to tell pip to freeze all the packages and their versions. I keep this file in a deploy folder in my project.

$ mkdir /srv/myproject.com/code/deploy
$ pip freeze > /srv/myproject.com/code/deploy/requirements.txt

Now, if I needed to recreate the virtual environment somewhere else, I could just tell pip to install all the packages from that file.

1	`$ pip install -r /srv/myproject.com/code/deploy/requirements.txt`

Configure uWSGI

Now that I have something to serve, I’ll configure uWSGI to serve it. The first step is to create a configuration file for the project. I call mine wsgi.py and store it in /srv/myproject.com/code/myproject/. It appends the current directory to the Python path, specifies the Django settings file for the project, and registers the WSGI handler.

import sys
import os

sys.path.append(os.path.abspath(os.path.dirname(__file__)))
os.environ['DJANGO_SETTINGS_MODULE'] = 'myproject.settings'

import django.core.handlers.wsgi

application = django.core.handlers.wsgi.WSGIHandler()

With that done, the next step is to decide how uWSGI should be run. I’m going to use Ubuntu’s upstart to supervise the service. I keep the upstart script in my project’s deploy/ directory.

$ vim /srv/myproject.com/code/deploy/uwsgi.conf

description "uWSGI server for My Project"

start on runlevel [2345]
stop on runlevel [!2345]

respawn
exec /usr/local/sbin/uwsgi \
--home /home/demo/.virtualenvs/myproject/ \
--socket /var/run/myproject.com.sock \
--chmod-socket \
--pythonpath /srv/myproject.com/code/ \
--module myproject.wsgi \
--process 2 \
--harakiri 30 \
--master \
--logto /srv/myproject.com/logs/uwsgi.log

Sadly, upstart doesn’t seem to recognize links. Rather than linking the config file into /etc/init/, I have to copy it.

$ sudo cp /srv/myproject.com/code/deploy/uwsgi.conf /etc/init/uwsgi-myproject.conf

Configure nginx

Nginx’s configuration is pretty straight-forward. If you’ve never configured the server before, Slicehost’s articles can set you down the right path. My own nginx config looks something like this:

user www-data www-data;
worker_processes 4;
pid /var/run/nginx.pid;

events {
        worker_connections 768;
        use epoll;
}

http {

    ##
    # Basic Settings
    ##

    sendfile on;
    tcp_nopush on;
    tcp_nodelay on;
    keepalive_timeout 30;
    types_hash_max_size 2048;
    # server_tokens off;

    # server_names_hash_bucket_size 64;
    # server_name_in_redirect off;

    include /etc/nginx/mime.types;
    default_type application/octet-stream;

    ##
    # Logging Settings
    ##

    access_log /var/log/nginx/access.log;
    error_log /var/log/nginx/error.log;

    ##
    # Gzip Settings
    ##

    gzip on;
    gzip_disable "msie6";
    gzip_types text/plain text/css application/json application/x-javascript text/xml application/xml application/xml+rss text/javascript;
    gzip_proxied any;
    gzip_comp_level 2;

    # gzip_vary on;
    # gzip_buffers 16 8k;
    # gzip_http_version 1.1;

    ##
    # Virtual Host Configs
    ##

    include /etc/nginx/conf.d/*.conf;
    include /etc/nginx/sites-enabled/*;
}

I keep the virtual host config for my project inside the project’s code/deploy/ directory. A basic virtual host for a Django project would looks like this:

server {
    listen 80;
    server_name www.myproject.com;
    rewrite ^/(.*) http://myproject.com/$1 permanent;
}

server {
    listen 80;
    server_name myproject.com;

    access_log /srv/myproject.com/logs/access.log;
    error_log /srv/myproject.com/logs/error.log;

    location /media {
        root /srv/myproject.com/public/;
    }

    location /static {
        root /srv/myproject.com/public/;
    }

    location / {
        uwsgi_pass unix:///var/run/myproject.com.sock;
        include uwsgi_params;
    }
}

To install and enable the virtual host, I’ll link the configuration file first to the nginx sites-available directory, and then link that link to the sites-enabled directory.

$ sudo ln -s /srv/myproject.com/code/deploy/nginx.conf /etc/nginx/sites-available/myproject.com
$ sudo ln -s /etc/nginx/sites-available/myproject.com /etc/nginx/sites-enabled/myproject.com

SSL

If you need to encrypt communications, Linode has a tutorial on using both self-signed certificates and commercial certificates with nginx.

Fire it Up

Nginx should be set to talk to uWSGI, which should be set to talk to the Django project. Time for a test run!

$ sudo start uwsgi-myproject
$ sudo /etc/init.d/nginx start

memcached

Django has a very good built-in cache framework. I like to take advantage of it with a memory-based backend: namely, memcached. It’s fast, efficient, and easy to setup.

All that’s needed is to install memcached on the server, followed by the Python API python-memcached.

$ sudo apt-get install memcached
$ workon myproject
$ pip install python-memcached

The default configuration file in Ubuntu lives at /etc/memcached.conf. I usually stick with the defaults, but sometimes end up changing the port that memchached runs on or the amount of memory it is allowed to use.

logrotate

With the web server more-or-less complete, I like to setup logrotate to manage the logs in my project’s directory. Once again, Slicehost has an excellent introduction to logrotate and an example config for virtual hosts.

I maintain a configuration file for each of the domains being served by the machine. The file for a domain lives in – you guessed it – the associated project’s deploy/ folder. Each contains two entries: one for the nginx virtual host and one for the uWSGI instance. The reason for this is that each config block needs a postrotate section to restart the associated server after the logs have been rotated. I don’t want nginx to be restarted everytime a uWSGI log is rotated, and I don’t want uWSGI restarted everytime an nginx log is rotated.

$ vim /srv/myproject.com/code/deploy/logrotate

/srv/myproject.com/logs/access.log /srv/myproject.com/logs/error.log {
    rotate 14
    daily
    compress
    delaycompress
    sharedscripts
    postrotate
        [ ! -f /var/run/nginx.pid ] || kill -USR1 `cat /var/run/nginx.pid`
    endscript
}

/srv/myproject.com/logs/uwsgi.log {
    rotate 14
    daily
    compress
    delaycompress
    postrotate
       restart --quiet uwsgi-myproject
    endscript
}

This file is linked in to the /etc/logrotate.d/ directory. Logrotate will automatically include any file in that directory inside its configuration.

$ sudo ln -s /srv/myproject.com/code/deploy/logrotate /etc/logrotate.d/myproject

Database Server

A web server isn’t much use without a database these days. I use PostgreSQL.

Install

1	`$ sudo apt-get install postgresql`

Configure

PostgreSQL has some unique terminology and ways of doing things. When I first set it up for the first time, having coming from a MySQL background, not everything was completely straightforward. As usual, Slicehost has a number of articles that will provide a foundation.

In the /etc/postgresql/8.4/main/postgresql.conf file, I uncomment the following two lines:

track_counts = on
autovacuum = on

Then restart the database server.

1	`$ sudo /etc/init.d/postgresql-8.4 restart`

After that I’ll change the password for the postgres user and the postgres database role.

$ sudo passwd postgres
$ sudo -u postgres psql
\password postgres
\q

To allow local socket connections to the database using passwords, I open up /etc/postgresql/8.4/main/pg_hba.conf and find the following line:

local   all         all                               ident

Which I then change to:

local   all         all                               md5

After which another restart is in order.

1	`$ sudo /etc/init.d/postgresql-8.4 restart`

Create a database

The next step is to create a user (or role, in PostgreSQL’s parlance) and database for the project. I use the same name for both.

$ sudo -u postgres createuser -PE myproject
$ sudo -u postgres createdb -O myproject myproject

After that, I should be able to connect.

1	`$ psql -U myproject`

Import the Database

If I’m restoring a previous database from a backup, now would be the time to import the backup.

1	`$ psql -U myproject < myproject.postgresql`

And now Django should be able to connect!

The basic server is setup and secure. Django, uWSGI, nginx and PostgreSQL are all running and getting along swimmingly. At this point, many people would be done, but I also like to have a minimal mail server.

Mail Server

Most of my domains use Google Apps, so I don’t need a full-blown mail server. I do want programs and scripts to be able to send mail, and I prefer not to do so through an external SMTP server – I’d rather just deal with having sendmail running on my own box. And I do have a few domains that do not use Google Apps. They have one or two aliases associated with them, so the server needs to receive messages for those domains and forward them off to an external address. If any of this sounds vaguely familiar, it’s because it’s the same thing I detailed last time. My setup now is the same as then, so I won’t repeat any of it here.

For a more detailed explanation of running Postfix, you can read the Slicehost articles.

A Note on Git

I use Git to keep track of the code for all my projects. (If you’re new to Git, you ought to skim the Git Reference or Everyday GIT With 20 Commands Or So). To manage websites, I create a repository of the directory with the code that runs the site (in this case, /srv/myproject.com/code/) and another empty, bare repository to work as a hub. With a post-update and post-commit, the end result is an excellent web workflow:

A copy of the hub can be checked out on a local machine for development. Whenever a change is committed, a simple git push will push the code to the web server and automatically make it live.
Changes can be made on the server in the actual live website directory. (This is not a best practice, but I do it more often than I should probably admit.) Whenever a change is committed, it is automatically pushed to the hub, so that a simple git pull is all that’s needed on the development machine to update its repository.

A more detailed explanation of this workflow is at Joe Maller’s blog.

To start, I need to create a repository for the new project I created in this tutorial. And, since this is a new server, I need to give Git my name and email address to record with every commit.

$ git config --global user.name "Pig Monkey"
$ git config --global user.email "pm@pig-monkey.com"
$ cd /srv/myproject.com/code/
$ git init

Before adding the files, I create a .gitignore file in the repository root to tell Git to ignore compiled Python files.

$ vim .gitignore

*.pyc

Now I add all the files to the repository, confirm that it worked, and commit the files.

$ git add .
$ git status -s
$ git commit -m "Initial commit of myproject.com"

I create the bare hub directory directly along side the projects code/.

$ cd ../
$ mkdir hub.git
$ cd hub.git
$ git --bare init

With the hub created, I need to add it as the remote for the main repository and push the master branch to it.

$ cd ../code
$ git remote add hub /srv/myproject.com/hub.git
$ git remote show hub
$ git push hub master

Now the hub needs a post-update script so that every time something is pushed to it, that change is automagically pulled into the live website directory.

$ vim /srv/myproject.com/hub.git/hooks/post-update

#!/bin/sh
echo
echo "**** Pulling changes into live"
echo

cd /srv/myproject.com/code || exit
unset GIT_DIR
git pull hub master

exec git-update-server-info

$ chmod +x /srv/myproject.com/hub.git/hooks/post-update

And the live website directory requires a post-commit script so that every time something is committed inside of it, that change is automagically pushed to the hub.

$ vim /srv/myproject.com/code/.git/hooks/post-commit

#!/bin/sh
echo
echo "**** pushing changes to Hub"
echo

git push hub

$ chmod +x /srv/myproject.com/code/.git/hooks/post-commit

All that’s left is to check out the hub onto the development machine – my laptop, in this case!

$ mkdir ~/work/myproject/
$ cd ~/work/myproject/
$ git clone ssh://myserver.com/srv/myproject.com/hub.git code

To test things out, we can add a file to the repository on the development machine.

$ cd code/
$ touch test
$ git add test
$ git commit -m "A test"
$ git push

Now go back to the server, and the file should be there! To test things the other way around, I’ll delete the file from the live repository.

$ cd /srv/myproject.com/code/
$ ls
myproject test
$ git rm test
$ git commit -m "Removing the test file"

And once again to the development machine:

$ git pull
$ ls
deploy  myproject

No more test! It’s pretty dandy.

Restoring

If I was building a new server and restoring a project from an old server, I would simply mirror the old hub and then clone that in the live directory.

$ cd /srv/myproject.com/
$ git clone --mirror ssh://myoldserver.com/srv/myproject.com/hub.git
$ git clone hub.git code/

Resources

Prior to building this server, I was new to a lot of this – particularly, uWSGI and virtualenv. The following tutorials helped me a good deal in putting together the perfect setup for my needs.

A Primer on virtualenv by Chris Scott
A web-focused Git workflow by Joe Maller
Deployment with uWSGI and nginx by Zachary Voase
Django on uWSGI and Nginx by Brandon Konkle
Django, Nginx and uWSGI in production by Jeremy Bowers
Notes on using pip and virtualenv with Django by Eliot
Presentation: pip and virtualenv by Rich Leland
Provisioning a new Ubuntu server for Django by Brandon Konkle
Running Django with Nginx and uWSGI by Simon Westphahl
virtualenvwrapper Command Reference
Working with virtualenv by Arthur Koziel

Back to Dreamhost

2009-05-22T00:00:00-07:00

I’ve moved this domain back to Dreamhost. It’s been a good run with Slicehost. I still think that they are the superior host (I’ve already noticed a slowness with Dreamhost compared to Slicehost) and I greatly prefer the freedom and control of a VPS over shared hosting, but Dreamhost is cheaper and I need to cut down expenses. In the future I may move back to Slicehost again.

If anybody finds anything broken due to the move, let me know.

Thoughts on SSH Security

2008-10-03T00:00:00-07:00

OpenSSH has a history of security. Only rarely are holes found in the actual program. It’s much more likely that a system will be compromised through poor configuration of the SSH daemon. Ideally, an SSH config would allow only protocol 2 connections, allow only specified users to connect (and certainly not root), disable X11 forwarding, disable password authentication (forcing ssh keys instead), and allowing connections only from specified IPs. These config options would look like this:

Protocol 2
PermitRootLogin no
AllowUsers demo
X11Forwarding no
PasswordAuthentication no

Allowing connections from only specified IP addresses would be accomplished by adding something like the following to /etc/hosts.deny:

sshd: ALL # Deny all by default
sshd: 192.168.1.0/255.255.255.0 # Allow this subnet
sshd: 4.2.2.1 # Allow this IP

(You could also accomplish this with iptables, but I think editing the above file is simpler.)

But the last two options – disabling password auth and allowing only certain IP addresses – limits mobility. I constantly login to my slice from multiple IPs, and I also need to login during travel when I may or may not have my key on me.

The main thing these two options protect against is a brute force attack. By allowing password logins from any IP, we give the attacker the ability to exploit the weakest part of SSH. This is where DenyHosts comes in.

DenyHosts is a python script which attempts to recognize and block brute force attacks. It has many attractive features and is included in the default Ubuntu repositories.

1	`$ sudo aptitude install denyhosts`

The config file is located at /etc/denyhosts.conf. It is very simply and readable. I recommend reading through it, but most of the default options are acceptable. If any changes are made, the daemon must be restarted:

1	`$ sudo /etc/init.d/denyhosts restart`

Default Ports

Many people also advocating changing SSH’s default port to something other than 22 (more specifically, something over 1024, which won’t be scanned by default by nmap). The argument in support of this is that many automated attack scripts look for SSH only on port 22. By changing the port, you save yourself the headache of dealing with script kiddies. Opponents to changing the port would argue that the annoyance of having to specify the port number whenever using ssh or scp outweighs the minute security benefits. It’s a heated argument. I lean toward leaving SSH on the default port.

An Ubuntu VPS on Slicehost: Basic Setup

2008-06-10T00:00:00-07:00

As mentioned previously, I’ve recently moved this domain over to Slicehost. What follows is Part One of a guide, compiled from my notes, to setting up an Ubuntu Hardy VPS. See also Part Two, Part Three, and Part Four.

Slicehost has an excellent article repository, containing guides on a number of subjects. After building a fresh Slice, you should first follow Part 1 and Part 2 of Slicehost’s basic setup articles.

I use slightly different coloring in my bash prompt, so, rather than what Slicehost suggests in their article, I add the following to ~/.bashrc:

export PS1='\[\033[0;32m\]\u@\[\033[0;35m\]\h\[\033[0;33m\] \w\[\033[00m\]: '

This is a good time to protect SSH by installing DenyHosts, which I discuss here:

1	`$ sudo aptitude install denyhosts`

Ubuntu’s default text editor is nano, which I abhor. Real men use vim. Ubuntu comes with a slimmed down version of vim, but you’ll probably want the full version:

1	`$ sudo aptitude install vim`

To change the global default editor variable, execute the following and select the editor of your choice:

1	`$ sudo update-alternatives --config editor`

This is also a perfect time to install GNU Screen.

1	`$ sudo aptitude install screen`

If you’re not familiar with Screen, Red Hat Magazine has a nice little introduction

My .screenrc looks like this:

# Print a pretty line at the bottom of the screen
hardstatus alwayslastline
hardstatus string '%{= kG}[ %{G}%H %{g}][%= %{=kw}%?%-Lw%?%{r}(%{W}%n*%f%t%?(%u)%?%{r})%{w}%?%+Lw%?%?%= %{g}][%{Y}%Y-%m-%d %{W}%c %{g}]'

# Nobody likes startup messages
startup_message off

# Turn visual bell on and set the message to display for only a fraction of a second
vbell on
vbellwait .3

# Set default shell title to blank
shelltitle ''

# Gimme my scrollback!
defscrollback 5000

# Change command character to backtick
escape ``

# Stop programs (like vim) from leaving their contents
# in the window after they exit
altscreen on

# Default screens
screen -t shell 0

I prefer to have my bash profile setup to connect me to Screen as soon as I login. If there are no running sessions, it will create one. If there is a current session, it will disconnect the session from wherever it is connected and connect it to my login. When I disconnect from screen, it automatically logs me out. To achieve this, I add the following to ~/.bashrc:

# If possible, reattach to an existing session and detach that session
# elsewhere. If not possible, create a new session.
if [ -z "$STY" ]; then
    exec screen -dR
fi

I would also recommend following Slicehost’s guide to installing chkrootkit and rkhunter.

One more thing: let’s set the timezone of the server to whatever is local to you (Slicehost’s Ubuntu image defaults to UTC). To do that, run:

1	`$ sudo dpkg-reconfigure tzdata`

Next up: install a web server.

An Ubuntu VPS on Slicehost: Mail

2008-06-10T00:00:00-07:00

As mentioned previously, I’ve recently moved this domain over to Slicehost. What follows is Part Three of a guide, compiled from my notes, to setting up an Ubuntu Hardy VPS. See also Part One, Part Two, and Part Four.

Last week I moved this domain’s email to Google Apps. Slicehost has a guide to creating MX records for Google Apps. I have a couple other domains with Google Apps, along with a couple domains hosted locally with addresses that simply forward to my primary, Google hosted, email. I also need to send mail from the server. To accomplish all of this, I use Postfix.

Installing Postfix is a simple matter. Telnet is used quite a bit for testing, so I install that too:

1	`$ sudo aptitude install postfix telnet mailutils`

The Postfix setup will ask how it should be installed – we want the “Internet Site” option – and then ask you for your fully qualified domain name.

Done? Let’s make sure Postfix is running:

1	`$ telnet localhost 25`

If it’s working Postfix should return:

Trying 127.0.0.1...
Connected to localhost.
Escape character is '^]'.
220 localhost ESMTP Postfix (Ubuntu)

Let’s send a test message from root to the user account user (replace that with whatever your standard user is):

ehlo localhost
mail from: root@localhost
rcpt to: user@localhost
data
Subject: Test
Hi, is this thing on?
.
quit

Now, check your email as user by running mail. See the message? Good.

Open /etc/postfix/main.cf to make sure that Postfix knows what domains it’s receiving mail for. To do this, edit the mydestination variable to include all the proper domains. For me, the name of my server looks like server.mydomain.com. I want Postfix to accept mail for that domain, but not for mydomain.com (since that’s being handled by Google Apps), so mine looks like:

mydestination = server.mydomain.com, localhost.mydomain.com , localhost

Restart Postfix if you made any changes:

1	`$ sudo /etc/init.d/postfix restart`

Right. Now let’s send another test. Notice this time we’re using full domain names, instead of localhost:

$ telnet server.mydomain.com 25

ehlo server.mydomain.com
mail from: root@server.mydomain.com
rcpt to: user@server.mydomain.com
data
Subject: domains!
woot... I think this works.
.
quit

Working? Good.

Let’s test from the outside. The first step is to open up the correct ports in the firewall. Assuming you have iptables configured in the way the Slicehost article suggests, open up your /etc/iptables.test.rules and add the following:

# Allow mail server connections
-A INPUT -p tcp -m state --state NEW --dport 25 -j ACCEPT

Now let’s apply the rules:

1	`$ sudo iptables-restore < /etc/iptables.test.rules`

Make sure everything looks dandy:

1	`$ sudo iptables -L`

If it meets your fancy, save the rules:

$ sudo -i
$ iptables-save > /etc/iptables.up.rules

And now, from your local computer, let’s test it out.

$ telnet server.mydomain.com 25

ehlo server.mydomain.com
mail from: root@server.mydomain.com
rcpt to: user@server.mydomain.com
data
Subject: remote connection test
Hello, you.
.
quit

Now check your mail on the mail server as before. Once again, everything should be working.

Now we need to setup a virtual domain. Remember, I don’t want any virtual users. I only want aliases at a virtual domain to forward to my primary email address. That makes this relatively simple. (Be very, very happy. You should have seen this guide before, when I was still hosting virtual domains with virtual users!) Open up /etc/postfix/main.cf and add the following:

virtual_alias_domains = myvirtualdomain.com
virtual_alias_maps = hash:/etc/postfix/virtual

Create the /etc/postfix/virtual file referenced above and add the aliases:

alias@myvirtualdomain.com       user@mydomain.com

Turn it into a database:

$ cd /etc/postfix
$ sudo postmap virtual

Restart Postfix:

1	`$ sudo /etc/init.d/postfix restart`

Attempt to send an email to the new alias at the virtual domain:

$ telnet server.mydomain.com 25
ehlo server.mydomain.com
mail from: root@server.mydomain.com
rcpt to: alias@myvirtualdomain.com
data
Subject: virtual domain test
I hope this works!
.
quit

The message should now be in your primary email inbox!

As long as we’re setting up forwards, let’s forward system account mail to somewhere where it’ll actually get read. To do so, create a ~/.forward file with the following contents:

user@mydomain.com

Let’s also create a /root/.forward, so that roots mail gets forwarded to my local account (where it is then forwarded to my primary email). Root’s forward would simply read:

user

Next up: install Wordpress with rewrites. (Previously, we did a basic setup and installed a web server.)

An Ubuntu VPS on Slicehost: Web Server

2008-06-10T00:00:00-07:00

As mentioned previously, I’ve recently moved this domain over to Slicehost. What follows is Part Two of a guide, compiled from my notes, to setting up an Ubuntu Hardy VPS. See also Part One, Part Three, Part Four.

Now we’ve got a properly configured, but idle, box. Let’s do something with it.

Nginx is a small, lightweight web server that’s all the rage on some small corners of the Net. Apache is extremely overkill for a small personal web server like this and, since we’re limited to 256MB of RAM on this VPS, it quickly becomes a resource hog. Lighttpd is another small, lightweight web server, but I’m a fan of Nginx. Try it out.

First, we need to install the web server. Nginx is now in Ubuntu’s repositories:

1	`$ sudo aptitude install nginx`

That’s all it takes in Hardy, but if you really want a guide for it, Slicehost has you covered.

Slicehost has a few more useful guides to Nginx, including introductions to the config layout and how to get started with vhosts:

Next up, we’ll need to install MySQL and PHP, and get them working with Nginx.

Slicehost has a guide for installing MySQL and Ruby on Rails, which also includes suggestions on optimizing MySQL. I follow the MySQL part of the guide, stopping at “Ruby on Rails install”.

Now MySQL is working, lets install PHP:

1	`$ sudo aptitude install php5-common php5-cgi php5-mysql php5-cli`

To get PHP as FastCGI working with Nginx, we first have to spawn the fcgi process. There are a few different ways to do that. Personally, I use the spawn-fcgi app from lighttpd. To use it, we’ll compile and make lighttpd, but not install it. We’re only after one binary.

Lighttpd has a few extra requirements, so let’s install those:

1	`$ sudo aptitude install libpcre3-dev libbz2-dev`

Now, download the source and compile lighttpd. Then copy the spawn-fcgi binary to /usr/bin/:

$ wget http://www.lighttpd.net/download/lighttpd-1.4.19.tar.gz
$ tar xvzf lighttpd-1.4.19.tar.gz
$ cd lighttpd-1.4.19
$ ./configure
$ make
$ sudo cp src/spawn-fcgi /usr/bin/spawn-fcgi

Then, create a script to launch spawn-fci (I call it /usr/bin/php5-fastcgi):

#!/bin/sh
/usr/bin/spawn-fcgi -a 127.0.0.1 -p 9000 -u www-data -C 2 -f /usr/bin/php5-cgi

The script tells spawn-fcgi to launch a fastcgi process, listening on 127.0.01:9000, owned by the web user, with only 2 child processes. You may want more child processes, but I’ve found 2 to be optimal.

Give the script permissions:

1	`$ sudo chmod +x /usr/bin/php5-fastcgi`

I then link the script filename to a version-neutral, err, version:

1	`$ sudo ln -s /usr/bin/php5-fastcgi /usr/bin/php-fastcgi`

Now we need an init script to start the process at boot. I use this one from HowToForge, named /etc/init.d/fastcgi:

#!/bin/bash
PHP_SCRIPT=/usr/bin/php-fastcgi
RETVAL=0
case "$1" in
    start)
        echo "Starting fastcgi"
        $PHP_SCRIPT
        RETVAL=$?
    ;;
stop)
        echo "Stopping fastcgi"
        killall -9 php5-cgi
        RETVAL=$?
    ;;
restart)
        echo "Restarting fastcgi"
        killall -9 php5-cgi
        $PHP_SCRIPT
        RETVAL=$?
    ;;
    *)
        echo "Usage: php-fastcgi {start|stop|restart}"
        exit 1
    ;;
esac      
exit $RETVAL

Give it permissions:

1	`$ sudo chmod 755 /etc/init.d/fastcgi`

Start it:

1	`$ sudo /etc/init.d/fastcgi start`

Have it start at boot:

1	`$ sudo update-rc.d fastcgi defaults`

Alright, now that PHP is running how we want it to, let’s tell Nginx to talk to it. To do that, add the following to your vhost server block in /etc/nginx/sites-available/mydomain.com, making sure to change the SCRIPT_FILENAME variable to match your directory structure:

location ~ \.php$ {
    fastcgi_pass   127.0.0.1:9000;
    fastcgi_index  index.php;
    fastcgi_param  SCRIPT_FILENAME  /home/user/public_html/mydomain.com/public$fastcgi_script_name;
    include        /etc/nginx/fastcgi.conf;
}

Now let’s create that /etc/nginx/fastcgi.conf file that’s being included above. As per the Nginx wiki article, mine looks like this:

fastcgi_param  GATEWAY_INTERFACE  CGI/1.1;
fastcgi_param  SERVER_SOFTWARE    nginx;
fastcgi_param  QUERY_STRING       $query_string;
fastcgi_param  REQUEST_METHOD     $request_method;
fastcgi_param  CONTENT_TYPE       $content_type;
fastcgi_param  CONTENT_LENGTH     $content_length;
fastcgi_param  REQUEST_URI        $request_uri;
fastcgi_param  DOCUMENT_URI       $document_uri;
fastcgi_param  DOCUMENT_ROOT      $document_root;
fastcgi_param  SERVER_PROTOCOL    $server_protocol;
fastcgi_param  REMOTE_ADDR        $remote_addr;
fastcgi_param  REMOTE_PORT        $remote_port;
fastcgi_param  SERVER_ADDR        $server_addr;
fastcgi_param  SERVER_PORT        $server_port;
fastcgi_param  SERVER_NAME        $server_name;

Then restart Nginx:

1	`$ sudo /etc/init.d/nginx restart`

Let’s create a file named test.php in your domain’s public root to see if everything is working. Inside, do something like printing phpinfo.

Go to http://mydomain.com/test.php. See it? Good. If you get “no input file specified” or somesuch, you broke something.

If you create an index.php, and delete any index.html or index.htm you might have, you’ll notice Nginx throws a 403 Forbidden error. To fix that, find the line in your vhost config (/etc/nginx/sites-available/mydomain.com) under the location / block that reads index index.html; and change it to index index.php index.html;. Then restart Nginx.

If you want SSL with your Nginx, Slicehost has a guide for generating the certificate and another guide for installing it.

You’ll want to install OpenSSL first:

1	`$ sudo aptitude install openssl`

There is one bug in the second guide. In the first server module listening on port 443, which forwards www.domain1.com to domain1.com, the rewrite rule specifies the http protocol. So, in effect, what that rule does is forward you from a secure domain to unsecure: https://www.domain1.com to http://domain1.com. We want it to forward to a secure domain. Simply change the rewrite rule like thus:

rewrite ^/(.*) https://domain1.com permanent;

Next up: install a mail server. (Previously, we did a basic setup.)

An Ubuntu VPS on Slicehost: Wordpress

2008-06-10T00:00:00-07:00

As mentioned previously, I’ve recently moved this domain over to Slicehost. What follows is Part Four of a guide, compiled from my notes, to setting up an Ubuntu Hardy VPS. See also Part One, Part Two, and Part Three.

I prefer to install Wordpress via Subversion, which makes updating easier. We’ll have to install Subversion on the server first:

1	`$ sudo aptitude install subversion`

After that, the Wordpress Codex has a guide to the rest of the install.

Nothing further is needed, unless you want fancy rewrites. In that case, we’ll have to make a change to your Nginx vhost config at /etc/nginx/sites-available/mydomain.com. Add the following to your server block under location / {:

# wordpress fancy rewrites
if (-f $request_filename) {
    break;
 }
 if (-d $request_filename) {
     break;
  }
  rewrite ^(.+)$ /index.php?q=$1 last;

While we’re here, I usually tell Nginx to cache static files by adding the following right above thelocation / { block:

# serve static files directly
location ~* ^.+\.(jpg|jpeg|gif|png|ico|zip|tgz|gz|rar|bz2|doc|xls|exe|pdf|ppt|txt|tar|mid|midi|wav|bmp|rtf|css)$ {
    root  /home/user/public_html/mydomain.com/public;
    expires 7d;
    break;
}

That’ll go in the https server section, too. Now, enable rewrites in your Wordpress config. I use the following “custom” structure:

/%year%/%monthnum%/%day%/%postname%/

Then, restart Nginx:

1	`$ sudo /etc/init.d/nginx restart`

And there you have it! You know have a working, new web server and mail server.

(Previously, we did a basic setup, installed a web server, and installed a mail server.)

A Move to Slicehost

2008-06-09T00:00:00-07:00

Yesterday I moved this domain over to Slicehost.

Ian first told me about Slicehost when we were both looking to move away from Dreamhost last November. Initially, we both intended to find another shared host, but that proved far too difficult – it seems most hosting companies have something against shared hosting with decent limits and ssh access (that last part is the kicker).

I signed up with Slicehost at the end of last year and tinkered around with it for a month or so, experimenting with setting up the server in different ways. Eventually, I found an Ubuntu-Nginx-PHP-MySQL-Postfix-Dovecot setup that I enjoyed, and one which I was comfortable administering. In the beginning of the year, I moved a couple of my domains over to the Slice. It’s been a great experience. I’m not sure why it took me 6 months to finally move this domain – my primary one – over. Running a VPS is deceivingly simple* and well worth the effort. If you’re currently running on a shared host and have some basic competency in a UNIX environment, I’d recommend giving it a shot.

In a bit I’ll post a series of guides, compiled from my notes, on how I setup the server.

It’s deceivingly simple if you’re not running a full mail server with virtual users running around everywhere. That part was a pain. Hence, the move to Google.