This combines the two most common store types, which modules often end up
opening with both interfaces separately anyway.
As well as combining them, I've taken the opportunity to improve some of the
method names to make them clearer.
Some of the OAuth stuff highlights a small need to retrieve a list of
roles somehow. Handy if you ever need a role selector in adhoc or
something.
Unless there's some O(n) thing we were avoiding?
On regular disconnects, </stream> is sent, then
sessionmanager.destroy_session() is called, then
sessionmanager.destroy_session() is called again when the TCP connection
is closed, from ondisconnect in mod_c2s.
It is a bit annoying and doesn't really tell you much.
Normally with bidi, any outgoing connection should be the same as the
incoming, hence when closing a bidi connection it should be removed as a
route to the remote server. However it is not guaranteed, a remote bidi-capable server
might have decided to open a new connection for some reason. This can
lead to a situation where there are two bidi connections, and the s2sout
route is a locally initiated s2sout connection. In this case, such a
s2sout connection should be kept.
Noticed in a rare case where bidi has just been enabled on a running
server, and something establishes new connections immediately when a
connection is closed.
The 'global' property should reflect whether the module API instance
represents the global context or a VirtualHost or Component context.
However the module:context() method did not override this, leading the
property of the previous module shining trough, leading to bugs in code
relying on the 'global' property.
See also #1736
This flag is something of a shortcut for `module.host == "*"` and should
always be equal to that. Its absence on the proxy object made the
property of the global module instance visible, causing problems such as
with URL reporting in mod_http
Since 5cd075ed4fd3 any file matching "fullchain" would be considered for
use.
Dehydrated stores fullchain certs in e.g, fullchain-1641171024.pem and a
symlink fullchain.pem pointing at the latest one. However the current
rule for finding a corresponding private key would try
privkey-1641171024.pem in the same directory, which may not exist.
The default config was updated in this way long ago, but if no option was
present in the config, Prosody would load internal_plain.
This change can result in changes (for the better) for people using very old
configuration files lacking an 'authentication' setting.
These changes to the API (hopefully the last) introduce a cleaner separation
between the user's primary (default) role, and their secondary (optional)
roles.
To keep the code sane and reduce complexity, a data migration is needed for
people using stored roles in 0.12. This can be performed with
prosodyctl mod_authz_internal migrate <host>
Goal: Introduce role-auth with minimal disruption
is_admin() is unsafe in a system with per-session permissions, so it has been
deprecated.
Roll-out approach:
1) First, log a warning when is_admin() is used. It should continue to
function normally, backed by the new role API. Nothing is really using
per-session authz yet, so there is minimal security concern.
The 'strict_deprecate_is_admin' global setting can be set to 'true' to
force a hard failure of is_admin() attempts (it will log an error and
always return false).
2) In some time (at least 1 week), but possibly longer depending on the number
of affected deployments: switch 'strict_deprecate_is_admin' to 'true' by
default. It can still be disabled for systems that need it.
3) Further in the future, before the next release, the option will be removed
and is_admin() will be permanently disabled.
Rationale:
- Removes a bunch of code!
- We don't have many cases where an actor is not bound to one of our hosts
- A notable exception is the admin shell, but if we ever attempt to lock those
sessions down, there is a load of other work that also has to be done. And
it's not clear if we would need a global authz provider for that anyway.
- Removes an extra edge case from the necessary mental model for operators
- Sessions that aren't bound to a host generally are anonymous or have an
alternative auth model (such as by IP addres).
- With the encapsulation now provided by util.roles, ad-hoc "detached roles"
can still be created anyway by code that needs them.
This commit was too awkward to split (hg record didn't like it), so:
- Switch to the new util.roles lib to provide a consistent representation of
a role object.
- Change API method from get_role_info() to get_role_by_name() (touches
sessionmanager and usermanager)
- Change get_roles() to get_user_roles(), take a username instead of a JID
This is more consistent with all other usermanager API methods.
- Support configuration of custom roles and permissions via the config file
(to be documented).
This is useful for a number of things. For example, listing users that need to
rotate their passwords after some event. It also provides a safer way for code
to determine that a user password has changed without needing to set a handler
for the password change event (which is a more fragile approach).
