You can try s3fs for Linux, or possibly wins3fs for Windows (though its last update is forever old).
There are also commercial alternatives like ExpanDrive, but these fail your “open source” criteria.
I would caution you to test whatever solution you pick rather carefully. I have used both open and proprietary approaches to “S3 as a file system,” and have never been satisfied with their reliability. Trying to couple what fundamentally is a remote, high-latency resource with a software system (the file system) that expects a very local, low-latency resource–well, the two often don’t match up very well. I’ve found that I can copy files beautifully to and from S3…and then a few minutes later, when I try to copy more files, the connection is somehow broken. This is an an environment with 50 Mbps network connections, in which software like ssh and scp have no problems whatever. My experience is that trying to treat S3 as a file system seemed like the very simplest and most fool-proof approach…but in practice, it often did not work robustly, and I spent more time restarting my S3 file system connection than it was worth. I have since switched to command-line tools like s3cmd which do not suffer the same drop-outs. YMMV.
