On Tue, Mar 16, 2010 at 19:06, Linas <linas_fi@ymail.com> wrote:
I had already this email draft in my head, but Ananda 'Arch Linux security is still poor' thread, on which the point was also brought up, moved me to really write it.
First off, there's an implicit level of trust on the package software, no matter which OS you use. When using Windows, you trust in Microsoft, when using Mac OS, you trust in Apple, when using a Linux distro, you trust the packagers and upstream. Either you do that or trust just whatever came installed and not install anything ever (thus not patching to new vulnerabilities).
The problem with Arch current packaging system is not that you must trust people able to write in core not to add a rm -rf / (to name the classical 'attack') nor that you didn't install arch with an infected media. The problem is that every time you do pacman -Syu, you must blindly trust that your dns, network, and mirror are reliable, too. The packages are verified with a md5 from the server list, but should you update from a compromised mirror (or impersonated, eg. arp poisoning, dns spoofing, bofh proxy operator...) you have lost. A pacman -Syu from an open wifi might be enough. A later update may 'clean' it, so you may not even notice that you were once compromised.
There are several ways to close the gap: *Always download the package list from ftp.archlinux.org It's the easier solution, but it only protects against the mirror operator. Moreover, it increases load on that server and makes it a single point of failure.
*Package lists are signed from a trusted master key. There may be up to a key per repo. Easy to provide, allows backward compatibility.
*Packages are automatically signed by ftp.archlinux.org before distributing them. Removes the dependancy over the package list. Packages can be shared securely (eg. getting a downgrade for an untrusted user).
*Each developer signs its own packages prior to uploading. Each repository key signs the keys of the developers with write access. Users can blacklist or trust independent developers.
Needless to say, the last solution is the one I like most. However, being more complete, it also means more work. :)
The package signing could be a simple text file with filenames and hashes (preferably something more secure than md5) signed with gpg, or could be expanded if more fields are needed.
Do you think this is a good idea? Which solution do you prefer? And most important, what would be needed to reach there?
__________________________________________________ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com
There is one last problem with trust that no one can cure. You either trust the devs or you don't. This is illustrated by a classic quote from Ken Thompson "The moral is obvious. You can't trust code that you did not totally create yourself. (Especially code from companies that employ people like me.) No amount of source-level verification or scrutiny will protect you from using untrusted code. In demonstrating the possibility of this kind of attack, I picked on the C compiler. I could have picked on any program-handling program such as an assembler, a loader, or even hardware microcode. As the level of program gets lower, these bugs will be harder and harder to detect. A well installed microcode bug will be almost impossible to detect." From this article http://cm.bell-labs.com/who/ken/trust.html .<http://cm.bell-labs.com/who/ken/trust.html> Myra Nelson