PuTTY wish ssh-fingerprint-formats

Home | FAQ | Feedback | Licence | Updates | Mirrors | Keys | Links | Team
Download: Stable · Snapshot | Docs | Changes | Wishlist

PuTTY currently only supports one format for displaying SSH public key fingerprints (used when verifying host keys); this is the traditional set of hex octets calculated with the MD5 hash function specified by section 4 of RFC 4716 (for instance ssh-rsa 1024 7b:e5:6f:a7:f4:f9:81:62:5c:e3:1f:bf:8b:57:6c:5a).

New versions of OpenSSH (from 6.8) default to displaying a different, base64-format fingerprint for certain keys, with current versions calculated using the SHA-256 hash (for instance SHA256:mVPwvezndPv/ARoIadVY98vAC0g+P/5633yTC4d/wXE).

Since modern OpenSSH takes some persuasion to display fingerprints in the old format, and we should be moving away from the weak MD5 hash anyway, PuTTY should support displaying fingerprints in this new format, as well as continuing to display the old format for backward compatibility.

We'd have to work out a UI for this. The obvious option is to display the fingerprint in all known formats in the host-key verification dialog; this will make the dialog even more intimidating, and could get unwieldy if we ever have more than two. A more complicated option would be to have a dropdown or buttons or something on the host-key dialog to let the user choose among the available formats; this probably implies a global setting to set the default (and PuTTY isn't currently very good at global settings).

(Should you need to get an MD5 fingerprint out of modern OpenSSH in the meantime, you need to run a command like

ssh-keygen -E md5 -l -f /etc/ssh/ssh_host_ed25519_key.pub

Some places where key fingerprints are currently used, so that extra UI complexity would need to be designed and implemented:

• The GUI dialog box (both Windows and GTK) asking for a host key confirmation. Obvious possibilities include a dropdown list of fingerprint types, or simply listing all the known ones in a huge info-dump. [We ended up listing SHA-256 by default, and putting other formats behind a "More info..." button.]
• The corresponding console message, which can't provide a dropdown list. Huge info-dump is one option; something you can type at the interactive prompt to provide a different fingerprint format is another [this is what we did].
• The readonly edit box displaying the fingerprint in Windows (GUI) PuTTYgen. The obvious approach here is to have an option in the menu bar that selects the format [so we did that].
• The GUI key list window in Windows Pageant. That window doesn't have a menu bar at the moment, but cluttering the window with a set of radio buttons seems a bit intrusive. Perhaps we should add a menu bar. [We added a drop-down.]
• The fingerprint output mode in Unix (command-line) PuTTYgen. Probably we do this one by making -O fingerprint and its synonym -l use the current default fingerprint style, and then add extra explicit modes like -O fingerprint-md5-hex or -O fingerprint-sha256-base64. [We added -E md5 instead.]
• The fingerprints in the output of pageant -l on Unix. A command-line option is the obvious way to select these. [We added --fptype.]
• The idiom on Unix Pageant's client command line that lets you select one of the currently loaded keys by typing a prefix of its fingerprint. [We added sha256: and md5: prefixes.]
• The configuration option in PuTTY itself which allows you to enter the fingerprint of a host key you want PuTTY to accept. We'd have to accept fingerprints in all formats, and do it in some way that ensured they were unambiguous (no fingerprint intended to be type X should be accepted on the grounds that it matched some key's type-Y fingerprint). [We added a SHA256: prefix; you can optionally specify MD5: too.]
• The diagnostic message in PuTTY's Event Log. That's text-only (and also goes to log files), so once a fingerprint has been written into it, it's too late to rewrite it in a different style. Perhaps that one should just print all available fingerprint types. [We just print SHA-256.]
• Diagnostic messages in Unix Pageant (only relevant in debug mode, so not vital). [Ditto.]

We have been dragging our feet about this, as the new OpenSSH scheme has some disadvantages, principally that it is harder to compare by hand and read out over the phone, as it makes no concession to the possibility of confusing homographs (letter-l versus number-1, O-vs-0 etc).

We did half-work-out an alternative scheme based on the the Uniform Data Fingerprint (UDF) scheme described in draft-hallambaker-udf. That scheme has been designed for readability while remaining relatively compact, to minimise the risk of ambiguity, and to allow easy migration to new hash algorithms (the first character encodes the algorithm); it would give fingerprints of the form MBQPX-76TZY-NXTKU-WJZIM-LH4T2-IYQU6. Our effort got as far as a half-written spec and some minimal proof-of-concept code, both unpublished. (However, the opportune moment to push this as a replacement for OpenSSH's scheme has clearly passed.)

2021-03-14: multiple formats are now supported, and the OpenSSH-style SHA256 fingerprint (SHA-256 of the standard public key blob, base64-encoded, with trailing = stripped) is now the default.