[snip] On Sat, Sep 14, 2013 at 10:36:55AM -0700, Don Armstrong wrote: > > To ensure that the checksums files themselves are correct, use GnuPG > to verify them against the accompanying signature files (e.g. > MD5SSUMS.sign). The keys used for these signatures are all in the > Debian GPG keyring and the best way to check them is to use that > keyring to validate via the web of trust. To make life easier for > users, here are the fingerprints for the keys that have been used > for releases in recent years (with some UIDs removed for clarity): >
Yes, and new users don't have that keyring. They might think verifying those fingerprints is safe. > > It would be fair to expect a large proportion of users cannot or will > > not be able to establish such a web of trust, especially if they're new > > users. > > Anyone capable of pulling off a MITM attack against an http site will be > capable of pulling off an attack against an https site too.[1] What is that claim based on? Breaking SSL/TLS is much more difficult than MITM-ing a plain http website, unless the signing CA is involved. And while I agree that MITM attacks don't happen really often for normal Internet connections, it's quite hazardous to assume http is safe when using Tor for example. > Even if the site were to be served via https, which is unlikely, due > to the fact that www.debian.org is mirrored, you still would have the > same set of problems, and users who were concerned about the > authenticity of the pages that they were viewing and the images that > they were downloading would still have to verify the key IDs via the > web of trust. Hm, I didn't know it's mirrored. Perhaps ftp-master should host those? Regardless, there has to be a website that's under Debian's control. > > 1: Unless you're proposing that people check the authenticity of the SSL > certificates too against a known set of fingerprints, which brings you > right back to the same bootstrapping problem. I'm proposing checking the SSL certs just like any browser checks them: verifying signatures against a set of trusted CAs. [snip] Also can we avoid placing this issue on the "wishlist" severity level? I don't mean to blow this out of proportion, but we're talking about a security issue regarding installing new Debian systems, after all. Regards, Eduard -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org