ToddGee.commie

For all of my DreamHost shell users, I've made all their passwords long, random passwords that are (near) impossible to hack.  This makes them near impossible to remember as well, so I use ssh public key authentication for all shell users.

Update: See 1st comment, below.

When we set up whip, our old workhorse server, a couple years ago, we disabled password based ssh logins, choosing to only allow access via public keys.  However, I occasionally needed the ability to access the system when I didn't have ready access to my private key (like when using someone else's / a public computer).  I came up with a system that would enable password-based ssh login for a period of time and then re-disable it.  This post describes that technique.  The system was actuated by a button an internal (private) web page (as described below.  On the same page, had a link to the MindTerm applet, also installed on the server.  This applet, which provides a terminal-like canvas with an ssh session, allowed me to access the server with only a java-enabled browser.

The system was run by a button press on a page.  The page was only accessible via a simple-auth challange (page served over SSL) to prevent shenanigans.  The button called a .cgi script that would, in turn, run, via sudo, anotheand r script that updated the sshd config (allowing password-based logins) and then restarted the sshd service.  This script also forked, the fork sleeping a set amount of time before restoring the original sshd config and (again) restarting sshd.  Note: all paths in these scripts (and related content) are specified with full paths for security.  (Additionally, everything is chmodded as unwritable.) It goes without saying that these paths were particular to the setup on whip...

16 characters:
http://www.pctools.com/guides/password/?length=16&phonetic=on&alpha=on&m...

Introduction

When I first signed up for DreamHost account, I understood, but wasn't overly fond of the way they ran web requests.

All requests to a deployed website on DreamHost are made in the context of the user owning the website.  That is, if a website is deployed under user 'X', calls into that website are performed as user 'X'.  This allows DreamHost to isolate security breaches to the user of the (defective) website.  However, this also has security implications for the individual websites.  If there is an exploitable defect within a particular website and a site is compromised,  files within that software could be modified as the call is done as the owning user.  Setting restrictive permissions (i.e. read-only) on the site's file helps, but doesn't completely mitigate the issue, as the files are owned by the calling user and (in a truly compromised environment) the file permissions could be modified.  In a traditional (non shared) web hosting system, this would not be an issue as web-initiated calls would be made by the (highly restricted) user (e.g apache, httpd, etc.) with permission to read (and not write) only those files required to serve the web page.

What follows is the scheme with which I came up to work around this issue.  In this discussion, I'll talk about Drupal sites; but the scheme is the same regardless of what files are deployed on a site.