Ability to read the program's manual from the terminal is much better
than to ask the user to search the manual page on the internet, we
already trust the installed program and documentation, but we should not
trust every manual page on the internet.
Updates happens multiple times, normally 2 to 3, even if we consider a
state without includes. On states with multiple includes, it could
easily get approximately 10 updates being ran. This behavior leads to
unnecessary network bandwidth being spent and more time to run the
installation state. When the connection is slow and not using the
cacher, such as torified connections on Whonix, the installation can
occurs much faster.
Adding external repositories has to be done prior to update to ensure it
is also fetched.
Fixes: https://github.com/ben-grande/qusal/issues/29
As it is not easy to get files to dom0 and we don't want to reimplement
a package manager, crude Git is the solution as of know.
With Git we have the following advantages: native fetch format for
source controlled files, cleaner command-line, automatic signature
verification during merge, the disadvantage is that it is not included
by default in Dom0 and filtering it's stdout chars are not possible.
Note that the remote can report messages to the client via stderr, which
is filtered already, and if it tries to send an escape sequence to
stdout, the operation will fail with 'bad line length character: CHAR'
printed to stderr on the client, unfiltered by qrexec, but filtered to
some extent by the git client. If it is an escape character, the char is
transformed to "?", but UTF-8 multibyte characters are not filtered. Up
to 4 bytes can be displayed.
Tar on the other hand is already installed, but it is much ancient and
it's file parsing caused CVEs in the past relatively more drastic than
Git, it also doesn't only include committed files, it can include any
file that is present in the directory, which by far, increases a lot of
the attack surface unless you reset the state to HEAD, clean .git
directory manually and there are possibly other avenues of attack.
Split-gpg V1 allowed for querying public keys, but as split-gpg2 is
running as an agent, public keys are not queried. Allowing connection to
the server to query only public parts of the key exposes the server more
than needed to the client.
All clients now have to hold the public key they need locally in order
to do GPG operations.
